You are on page 1of 287

Introduction to Complex Analysis

Michael Taylor

Contents

0.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.

Complex numbers, power series, and exponentials

Holomorphic functions, derivatives, and path integrals
Holomorphic functions defined by power series
Exponential and trigonometric functions: Eulers formula
Square roots, logs, and other inverse functions
The Cauchy integral theorem and the Cauchy integral formula
The maximum principle, Liouvilles theorem, and the fundamental theorem of algebra
Harmonic functions on planar regions
Moreras theorem and the Schwarz reflection principle
Goursats theorem
Uniqueness and analytic continuation
Singularities
Laurent series
Fourier series and the Poisson integral
Fourier transforms
Laplace transforms
Residue calculus
The argument principle
The Gamma function
The Riemann zeta function
Covering maps and inverse functions
Normal families
Conformal maps
The Riemann mapping theorem
Boundary behavior of conformal maps
The disk covers C \ {0, 1}
The Riemann sphere and other Riemann surfaces
Montels theorem
Picards theorems
Harmonic functions again: Harnack estimates and more Liouville theorems
Periodic and doubly periodic functions - infinite series representations
The Weierstrass in elliptic function theory
Theta functions and
Elliptic integrals
p
The Riemann surface of q()

A.
B.
C.
D.

Metric spaces, convergence, and compactness

Derivatives and diffeomorphisms
Surfaces and metric tensors
Greens theorem

E.
F.
G.
H.
I.
J.
K.

Poincare metrics
The fundamental theorem of algebra (elementary proof)
The Weierstrass approximation theorem
Inner product spaces
2 is irrational
Eulers constant
Rapid evaluation of the Weierstrass -function

Introduction
This text covers material presented in complex analysis courses I have taught numerous
times at UNC. The core idea of complex analysis is that all the basic functions that arise
in calculus, first derived as functions of a real variable, such as powers and fractional
powers, exponentials and logs, trigonometric functions and their inverses, and also a host
of more sophisticated functions, are actually naturally defined for complex arguments, and
are complex-differentiable (a.k.a. holomorphic). Furthermore, the study of these functions
on the complex plane reveals their structure more truly and deeply than one could imagine
by only thinking of them as defined for real arguments.
An introductory 0 defines the algebraic operations on complex numbers, say z = x + iy
and w = u + iv, discusses the magnitude |z| of z, defines convergence of infinite sequences
and series, and derives some basic facts about power series
(i.1)

f (z) =

ak z k ,

k=0

such as the fact that if this converges for z = z0 , then it converges absolutelyPfor |z| < R =
|z0 |, to a continuous function. It is also shown that, for z = t real, f 0 (t) = k1 kak tk1 ,
for R < t < R. Here we allow ak C. We define the exponential function
(i.2)

X
1 k
e =
z ,
k!
z

k=0

and use these observations to deduce that, whenever a C,

(i.3)

d at
e = aeat .
dt

We use this differential equation to derive further properties of the exponential function.
While 0 develops calculus for complex valued functions of a real variable, 1 introduces
calculus for complex valued functions of a complex variable. We define the notion of complex differentiability. Given an open set C, we say a function f : C is holomorphic
on provided it is complex differentiable, with derivative f 0 (z), and f 0 is continuous on .
Writing f (z) = u(z) + iv(z), we discuss the Cauchy-Riemann equations for u and v. We
also introduce the path integral and provide some versions of the fundamental theorem of
calculus in the complex setting. (More definitive results will be given in 5.)
In 2 we return to convergent power series and show they produce holomorphic functions.
We extend results of 0 from functions of a real variable to functions of a complex variable.
Section 3 returns to the exponential function ez , defined above. We extend (i.3) to
(i.4)

d az
e = aeaz .
dz

We show that t 7 et maps R one-to-one and onto (0, ), and define the logarithm on
(0, ), as its inverse:
(i.5)

x = et t = log x.

We also examine the behavior of (t) = eit , for t R, showing that this is a unit-speed
curve tracing out the unit circle. From this we deduce Eulers formula,
(i.6)

This leads to a direct, self-contained treatment of the trigonometric functions.

In 4 we discuss inverses to holomorphic functions. In particular, we extend the logarithm from (0, ) to C \ (, 0], as a holomorphic function. We define fractional powers
(i.7)

z a = ea log z ,

a C, z C \ (, 0],

and investigate their basic properties. We also discuss inverse trigonometric functions in
the complex plane.
In 5 we introduce a major theoretical tool of complex analysis, the Cauchy integral
theorem. We provide a couple of proofs, one using Greens theorem and one based simply
on the chain rule and the fundamental theorem of calculus. Cauchys integral theorem
leads to Cauchys integral formula, and then to the general development of holomorphic
functions on a domain C in power series about any p , convergent in any disk
centered at p and contained in .
Results of 5 are applied in 6 to prove a maximum principle for holomorphic functions,
and also a result called Liouvilles theorem, stating that a holomorphic function on C that
is bounded must be constant. We show that each of these results imply the fundamental
theorem of algebra, that every non-constant polynomial p(z) must vanish somewhere in C.
In 7 we discuss harmonic functions on planar regions, and their relationship to holomorphic functions.
In 8 we establish Moreras theorem, a sort of converse to Cauchys integral theorem.
We use this in 9 to prove Goursats theorem, to the effect that the C 1 hypothesis can be
dropped in the characterization of holomorphic functions.
After a study of the zeros and isolated singularities of holomorphic functions in 10
11, we look at other infinite series developments of functions: Laurent series in 12 and
Fourier series in 13. The method we use to prove a Fourier inversion formula also produces
a Poisson integral formula for the unique harmonic function in a disk with given boundary
values. Variants of Fourier series include the Fourier transform and the Laplace transform,
discussed in 1415. These transforms provide some interesting examples of integrals
whose evaluations cannot be done with the techniques of elementary calculus.
Residue calculus, studied in 16, provides a powerful tool for the evaluation of many definite integrals. A related tool with many important applications is the argument principle,
studied in 17. In a sense these two sections lie at the heart of the course.
In 18 and 19 we make use of many of the techniques developed up to this point to
study two special functions, the Gamma function and the Riemann zeta function. In both

cases these functions are initially defined in a half-plane and then analytically continued
as meromorphic functions on C.
Sections 2028 have a much more geometrical flavor than the preceding sections. We
look upon holomorphic diffeomorphisms as conformal maps. We are interested in holomorphic covering maps and in implications of their existence. We also study the Riemann
sphere, as a conformal compactification of the complex plane. We include two gems of
nineteenth century analysis, the Riemann mapping theorem and Picards theorem. An
important tool is the theory of normal families, studied in 21.
In 29 we return to a more function-theoretic point of view. We establish Harnack
estimates for harmonic functions and use them to obtain Liouville theorems of a more
general nature than obtained in 7. These tools help us produce some concrete illustrations
of Picards theorem, such as the fact that ez z takes on each complex value infinitely
often.
In 3033 we provide an introduction to doubly periodic meromorphic functions, also
known as elliptic functions, and their connection to theta functions and elliptic integrals,
and in 34 we show how constructions of compact Riemann surfaces provide tools in elliptic
function theory.
This text concludes with several appendices. In Appendix A we collect material on
metric spaces and compactness, including particularly the Arzela-Ascoli theorem, which is
an important ingredient in the theory of normal families. We also prove the contraction
mapping theorem, of use in Appendix B. In Appendix B we discuss the derivative of
a function of several real variables and prove the Inverse Function Theorem, in the real
context, which is used in 4 to get the Inverse Function Theorem for holomorphic functions
on domains in C.
Appendix C discusses metric tensors on surfaces. It is of use in 26, on the Riemann
sphere and other Riemann surfaces. This material is also useful for Appendix E, which
introduces a special metric tensor, called the Poincare metric, on the unit disk and other
domains in C, and discusses some connections with complex function theory, including
another proof of Picards big theorem. In between, Appendix D proves Greens theorem
for planar domains, of use in one proof of the Cauchy integral theorem in 5.
In Appendix F we give a proof of the Fundamental Theorem of Algebra that is somewhat
different from that given in 7. It is elementary, in the sense that it does not rely on
results from integral calculus.
In Appendix G we show that a construction arising in 14 to prove the Fourier inversion
formula also helps establish a classical result of Weierstrass on approximating continuous
functions by polynomials. In fact, this proof is basically that of Weierstrass, and I prefer
it to other proofs that manage to avoid complex function theory.
Appendix H deals with inner product spaces, and presents some results of use in our
treatments of Fourier series and the Fourier transform, in 1314. In Appendix I, we
present a proof that 2 is irrational. Appendix J has material on Eulers constant. Appendix K complements 32 with a description of how to evaluate (z) rapidly via the use
of theta functions.
Acknowledgment

Thanks to Shrawan Kumar for testing this text in his Complex Analysis course, for pointing
out corrections, and for other valuable advice.

0. Complex numbers, power series, and exponentials

A complex number has the form
(0.1)

z = x + iy,

where x and y are real numbers. These numbers make up the complex plane, which is
just the xy-plane with the real line forming the horizontal axis and the real multiples of i
forming the vertical axis. See Figure 0.1. We write
(0.2)

x = Re z,

y = Im z.

We write x, y R and z C. We identify x R with x + i0 C. If also w = u + iv with

u, v R, we have addition and multiplication, given by
z + w = (x + u) + i(y + v),

(0.3)

i2 = 1.

(0.4)

(0.5)

z + w = w + z,

zw = wz,

the associative laws (with also c C)

(0.6)

z + (w + c) = (z + w) + c,

z(wc) = (zw)c,

and the distributive law

(0.7)

c(z + w) = cz + cw,

as following from their counterparts for real numbers. If c 6= 0, we can perform division
by c,
z
= w z = wc.
c
See (0.13) for a neat formula.
For z = x+iy, we define |z| to be the distance of z from the origin 0, via the Pythagorean
theorem:
p
(0.8)
|z| = x2 + y 2 .

Note that
(0.9)

|z|2 = z z,

where z = x iy,

(0.10)

z + z = 2 Re z,

z z = 2i Im z,

and
(0.11)

z + w = z + w,

zw = z w.

(0.12)

We also have, for c 6= 0,

z
1
= 2 zc.
c
|c|

(0.13)

The following result is known as the triangle inequality, as Figure 0.2 suggests.
Proposition 0.1. Given z, w C,
(0.14)

|z + w| |z| + |w|.

Proof. We compare the squares of the two sides:

|z + w|2 = (z + w)(z + w)
(0.15)

= zz + ww + zw + wz
= |z|2 + |w|2 + 2 Re(zw),

while
(0.16)

(|z| + |w|)2 = |z|2 + |w|2 + 2|z| |w|

= |z|2 + |w|2 + 2|zw|.

Thus (0.14) follows from the inequality Re(zw) |zw|, which in turn is immediate from
the definition (0.8). (For any C, Re ||.)
We can define convergence of a sequence (zn ) in C as follows. We say
(0.17)

10

the latter notion involving convergence of a sequence of real numbers. Clearly if zn =

xn + iyn and z = x + iy, with xn , yn , x, y R, then
(0.18)

One readily verifies that

(0.19)

zn z, wn w = zn + wn z + w and zn wn zw,

as a consequence of their counterparts for sequences of real numbers.

A related notion is that a sequence (zn ) in C is Cauchy if and only if
(0.19A)

|zn zm | 0 as m, n .

As in (0.18), this holds if and only if (xn ) and (yn ) are Cauchy in R. The following is an
important fact.
(0.19B)

(0.19C)

each Cauchy sequence in R converges.

A detailed presentation of the field R of real numbers, including a proof of (0.19C), is given
in Chapter 1 of [T0].
We can define the notion of convergence of an infinite series
(0.20)

zk

k=0

(0.21)

sn =

n
X

zk .

k=0

(0.22)

sn w =

zk = w.

k=0

Note that
n+m

|sn+m sn | =
zk
k=n+1

(0.23)

n+m
X
k=n+1

|zk |.

11

(0.24)

|zk | < ,

k=0

i.e., there exists A < such that

N
X

(0.24A)

|zk | A,

N.

k=0

Then the sequence (sn ) given by (0.21) is Cauchy, hence the series (0.20) is convergent.
Proof. If (sn ) is not Cauchy, there exist a > 0, n % , and m > 0 such that
|sn +m sn | a.
Passing to a subsequence, one can assume that n+1 > n + m . Then (0.23) implies
mX
+n

|zk | a,

k=0

If (0.24) holds, we say the series (0.20) is absolutely convergent.
An important class of infinite series is the class of power series

(0.25)

ak z k ,

k=0

with ak C. Note that if z1 6= 0 and (0.25) converges for z = z1 , then there exists C <
such that
|ak z1k | C,

(0.25A)

k.

Hence, if |z| r|z1 |, r < 1, we have

(0.26)

X
k=0

|ak z | C

X
k=0

rk =

C
< ,
1r

the last identity being the classical geometric series computation. This yields the following.

12

Proposition 0.2. If (0.25) converges for some z1 6= 0, then either this series is absolutely
convergent for all z C, or there is some R (0, ) such that the series is absolutely
convergent for |z| < R and divergent for |z| > R.
We call R the radius of convergence of (0.25). In case of convergence for all z, we say
the radius of convergence is infinite. If R > 0 and (0.25) converges for |z| < R, it defines
a function
(0.27)

f (z) =

ak z k ,

z DR ,

k=0

(0.28)

Proposition 0.3. If the series (0.27) converges in DR , then f is continuous on DR , i.e.,

given zn , z DR ,
(0.29)

zn z = f (zn ) f (z).

Proof. For each z DR , there exists S < R such that z DS , so it suffices to show that
f is continuous on DS whenever 0 < S < R. Pick T such that S < T < R. We know that
there exists C < such that |ak T k | C for all k. Hence
z DS = |ak z k | C

(0.30)

S k
T

For each N , write

f (z) = SN (z) + RN (z),
(0.31)

SN (z) =

N
X

ak z k ,

RN (z) =

k=0

ak z k .

k=N +1

Each SN (z) is a polynomial in z, and it follows readily from (0.19) that SN is continuous.
Meanwhile,
(0.32)

z DS |RN (z)|

X
k=N +1

k
X
S
|ak z | C
= CN ,
T
k

k=N +1

and N 0 as N , independently of z DS . Continuity of f on DS follows, as a

consequence of the next lemma.

13

Lemma 0.3A. Let SN : DS C be continuous functions. Assume f : DS C and

SN f uniformly on DS , i.e.,
(0.32A)

|SN (z) f (z)| N , z DS ,

N 0, as N .

Then f is continuous on DS .
Proof. Let zn z in DS . We need to show that, given > 0, there exists M = M () <
such that
|f (z) f (zn )| , n M.
To get this, pick N such that (0.32A) holds with N = /3. Now use continuity of SN , to
deduce that there exists M such that
|SN (z) SN (zn )|

,
3

n M.

It follows that, for n M ,

|f (z) f (zn )| |f (z) SN (z)| + |SN (z) SN (zn )| + |SN (zn ) f (zn )|

+ + ,
3 3 3
as desired.
Remark. The estimate (0.32) says the series (0.27) converges uniformly on DS , for each
S < R.
A major consequence of material developed in 15 will be that a function on DR
is given by a convergent power series (0.27) if and only if f has the property of being
holomorphic on DR (a property that is defined in 1). We will be doing differential and
integral calculus on such functions. In this preliminary section, we restrict z to be real,
and do some calculus, starting with the following.
Proposition 0.4. Assume ak C and
(0.33)

f (t) =

a k tk

k=0

converges for real t satisfying |t| < R. Then f is differentiable on the interval R < t < R,
and
(0.34)

f (t) =

kak tk1 ,

k=1

the latter series being absolutely convergent for |t| < R.

14

We first check absolute convergence of the series (0.34). Let S < T < R. Convergence
of (0.33) implies there exists C < such that
|ak |T k C,

(0.35)

k.

Hence, if |t| S,
|kak tk1 |

(0.36)

C S k
k
,
S
T

which readily yields absolute convergence. (See Exercise 3 below.) Hence

(0.37)

g(t) =

kak tk1

k=1

is continuous on (R, R). To show that f 0 (t) = g(t), by the fundamental theorem of
calculus, it is equivalent to show
Z

(0.38)

0

The following result implies this.

Proposition 0.5. Assume bk C and
(0.39)

g(t) =

bk tk

k=0

converges for real t, satisfying |t| < R. Then, for |t| < R,
Z
(0.40)
0

X
bk k+1
g(s) ds =
t
,
k+1
k=0

the series being absolutely convergent for |t| < R.

Proof. Since, for |t| < R,
b

k k+1
t

R|bk tk |,
k+1

(0.41)

convergence of the series in (0.40) is clear. Next, parallel to (0.31), write

g(t) = SN (t) + RN (t),
(0.42)

SN (t) =

N
X
k=0

bk t ,

RN (t) =

X
k=N +1

bk tk .

15

(0.43)

|t| S |RN (t)| CN 0 as N ,

so
Z

(0.44)
0

Z t
N
X
bk k+1
g(s) ds =
t
+
RN (s) ds,
k+1
0
k=0

and
(0.45)

Z t
Z t

RN (s) ds
|RN (s)| ds CRN .

This gives (0.40).

We use Proposition 0.4 to solve some basic differential equations, starting with
(0.46)

f 0 (t) = f (t),

f (0) = 1.

We look for a solution as a power series, of the form (0.33). If there is a solution of this
form, (0.34) requires
(0.47)

a0 = 1,

ak+1 =

ak
,
k+1

i.e., ak = 1/k!, where k! = k(k 1) 2 1. We deduce that (0.46) is solved by

(0.48)

X
1 k
f (t) = e =
t ,
k!
t

t R.

k=0

This defines the exponential function et . Convergence for all t follows from the ratio test.
(Cf. Exercise 4 below.) More generally, we define
(0.49)

X
1 k
e =
z ,
k!
z

z C.

k=0

Again the ratio test shows that this series is absolutely convergent for all z C. Another
application of Proposition 0.4 shows that
(0.50)

eat =

X
ak
k=0

k!

tk

solves
(0.51)

d at
e = aeat ,
dt

16

whenever a C.
We claim that eat is the only solution to
f 0 (t) = af (t),

(0.52)

f (0) = 1.

To see this, compute the derivative of eat f (t):

d at
e f (t) = aeat f (t) + eat af (t) = 0,
dt

(0.53)

where we use the product rule, (0.51) (with a replaced by a), and (0.52). Thus eat f (t)
is independent of t. Evaluating at t = 0 gives
eat f (t) = 1,

(0.54)

t R,

whenever f (t) solves (0.52). Since eat solves (0.52), we have eat eat = 1, hence
eat =

(0.55)

1
,
eat

t R, a C.

Thus multiplying both sides of (0.54) by eat gives the asserted uniqueness:
f (t) = eat ,

(0.56)

t R.

We can draw further useful conclusions by applying d/dt to products of exponentials.

Let a, b C. Then

(0.57)

d at bt (a+b)t
e e e
dt
= aeat ebt e(a+b)t beat ebt e(a+b)t + (a + b)eat ebt e(a+b)t
= 0,

gives
(0.58)

t R.

(0.59)

e(a+b)t = eat ebt ,

t R, a, b C,

or, setting t = 1,
(0.60)

ea+b = ea eb ,

a, b C.

17

We will resume study of the exponential function in 3, and derive further important
properties.

Exercises
1. Supplement (0.19) with the following result. Assume there exists A > 0 such that
|zn | A for all n. Then
(0.61)

2. Letting sn =

zn z =
Pn
k=0

rk , write the series for rsn and show that

(1 r)sn = 1 rn+1 ,

(0.62)

1
1
.
zn
z

hence sn =

1 rn+1
.
1r

Deduce that
(0.63)

0 < r < 1 = sn

1
, as n ,
1r

as stated in (0.26).
3. The absolute convergence said to follow from (0.36) can be stated as follows:
(0.64)

0 < r < 1 =

krk is absolutely convergent.

k=1

Prove this.
Hint. Writing r = s2 , 0 < s < 1, deduce (0.64) from the assertion
(0.65)

Note that this is equivalent to

(0.66)

a > 0 =

k
is bounded, for k N.
(1 + a)k

Show that
(0.67)

(1 + a)k = (1 + a) (1 + a) 1 + ka,

Use this to prove (0.66), hence (0.65), hence (0.64).

a > 0, k N.

18

4. This exercise discusses the ratio test, mentioned in connection with the infinite series
(0.49). Consider the infinite series

(0.69)

ak ,

ak C.

k=0

a

k+1
(0.70)
k N =
r.
ak
Show that

(0.71)

|ak | < .

k=0

Hint. Show that

(0.72)

|ak | |aN |

k=N

r` =

`=0

|aN |
.
1r

5. In case
(0.73)

ak =

zk
,
k!

show that for each z C, there exists N < such that (0.70) holds, with r = 1/2.
6. This exercise discusses the integral test for absolute convergence of an infinite series,
which goes as follows. Let f be a positive, monotonically decreasing, continuous function
on [0, ), and suppose |ak | = f (k). Then

Z
|ak | <

N
X
k=1

f (t) dt < .
0

k=0

Prove this.
Hint. Use

Z
|ak |

f (t) dt
0

N
1
X
k=0

7. Use the integral test to show that, if a > 0,

X
1
< a > 1.
a
n
n=1

|ak |.

19

8. This exercise deals with alternating series. Assume bk & 0. Show that

(1)k bk is convergent,

k=0

n+m

(1)k bk bn .

k=n

9. Show that

k=1 (1)

10. Show that if f, g : (a, b) C are differentiable, then

d
f (t)g(t) = f 0 (t)g(t) + f (t)g 0 (t).
dt

(0.74)

Note the use of this identity in (0.53) and (0.57).

11. Use the results of Exercise 10 to show, by induction on k, that
d k
t = ktk1 ,
dt

(0.75)

k = 1, 2, 3, . . . ,

hence
Z

(0.76)
0

sk ds =

1 k+1
t
,
k+1

k = 0, 1, 2, . . . .

Note the use of these identities in (0.44), leading to many of the identities in (0.34)(0.51).

20

1. Holomorphic functions, derivatives, and path integrals

Let C be open, i.e., if z0 , there exists > 0 such that D (z0 ) = {z C :
|z z0 | < } is contained in . Let f : C. If z , we say f is complex-differentiable
at z, with derivative f 0 (z) = a, if and only if
(1.1)

lim

h0

1
[f (z + h) f (z)] = a.
h

Here, h = h1 + ih2 , with h1 , h2 R, and h 0 means h1 0 and h2 0. Note that

(1.2)

lim

h1 0

1
f
[f (z + h1 ) f (z)] =
(z),
h1
x

and
(1.3)

lim

h2 0

1
1 f
[f (z + ih2 ) f (z)] =
(z),
ih2
i y

provided these limits exist.

As a first set of examples, we have
1
[f (z + h) f (z)] = 1,
h
1
h
f (z) = z = [f (z + h) f (z)] = .
h
h
f (z) = z =
(1.4)

In the first case, the limit exists and we have f 0 (z) = 1 for all z. In the second case, the
limit does not exist. The function f (z) = z is not complex-differentiable.
Definition. A function f : C is holomorphic if and only if it is complex-differentiable
and f 0 is continuous on .
Adding the hypothesis that f 0 is continuous makes for a convenient presentation of the
basic results. In 9 it will be shown that every complex differentiable function has this
So far, we have seen that f1 (z) = z is holomorphic. We produce more examples of
holomorphic functions. For starters, we claim that fk (z) = z k is holomorphic on C for
each k Z+ , and
(1.5)

d k
z = kz k1 .
dz

21

Proposition 1.1. If f and g are holomorphic on , so is f g, and

(1.6)

d
(f g)(z) = f 0 (z)g(z) + f (z)g 0 (z).
dz

Proof. Just as in beginning calculus, we have

(1.7)

1
[f (z + h)g(z + h) f (z)g(z)]
h
1
= [f (z + h)g(z + h) f (z)g(z + h) + f (z)g(z + h) f (z)g(z)]
h
1
1
= [f (z + h) f (z)]g(z + h) + f (z) [g(z + h) g(z)].
h
h

The first term in the last line tends to f 0 (z)g(z), and the second term tends to f (z)g 0 (z),
as h 0. This gives (1.6). If f 0 and g 0 are continuous, the right side of (1.6) is also
continuous, so f g is holomorphic.
It is even easier to see that the sum of two holomorphic functions is holomorphic, and
(1.8)

d
(f (z) + g(z)) = f 0 (z) + g 0 (z),
dz

Hence every polynomial p(z) = an z n + + a1 z + a0 is holomorphic on C.

We next show that f1 (z) = 1/z is holomorphic on C \ 0, with
(1.9)

d 1
1
= 2.
dz z
z

In fact,
(1.10)

1h 1
1i
1
h
1

=
=
,
h z+h z
h z(z + h)
z(z + h)

which tends to 1/z 2 as h 0, if z 6= 0, and this gives (1.9). Continuity on C \ 0 is

readily established. From here, we can apply Proposition 1.1 inductively and see that z k
is holomorphic on C \ 0 for k = 2, 3, . . . , and (1.5) holds on C \ 0 for such k.
Next, recall the exponential function
(1.11)

X
1 k
e =
z ,
k!
z

k=0

(1.12)

d z
e = ez .
dz

22

ez+h = ez eh .

(1.13)
Hence
(1.14)

1 z+h
eh 1
[e
ez ] = ez
.
h
h

Now (1.11) implies

(1.15)

k=1

k=0

eh 1 X 1 k1 X
1
=
h
=
hk ,
h
k!
(k + 1)!

and hence
(1.16)

eh 1
= 1.
lim
h0
h

This gives (1.12). Another proof of (1.12) will follow from the results of 2.
We next establish a chain rule for holomorphic functions. In preparation for this,
we note that the definition (1.1) of complex differentiability is equivalent to the condition
that, for h sufficiently small,
(1.17)

with
(1.18)

lim

h0

r(z, h)
= 0,
h

i.e., r(z, h) 0 faster than h. We write

(1.18)

r(z, h) = o(|h|).

Here is the chain rule.

Proposition 1.2. Let , O C be open. If f : C and g : O are holomorphic,
then f g : O C, given by
(1.19)

f g(z) = f (g(z)),

is holomorphic, and
(1.20)

d
f (g(z)) = f 0 (g(z))g 0 (z).
dz

23

(1.21)

with r(z, h) = o(|h|). Hence

f (g(z + h)) = f (g(z) + g 0 (z)h + r(z, h))
(1.22)

= f (g(z)) + f 0 (g(z))(g 0 (z)h + r(z, h)) + r2 (z, h)

= f (g(z)) + f 0 (g(z))g 0 (z)h + r3 (z, h),

(1.23)

r3 (z, h) = f 0 (g(z))r(z, h) + r2 (z, h) = o(|h|).

This implies f g is complex-differentiable and gives (1.20). Since the right side of (1.20)
is continuous, f g is seen to be holomorphic.
Combining Proposition 1.2 with (1.9), we have the following.
Proposition 1.3. If f : C is holomorphic, then 1/f is holomorphic on \ S, where
(1.24)

S = {z : f (z) = 0},

and, on \ S,
(1.25)

d 1
f 0 (z)
=
.
dz f (z)
f (z)2

We can also combine Proposition 1.2 with (1.12) and get

(1.26)

d f (z)
e
= f 0 (z)ef (z) .
dz

We next examine implications of (1.2)(1.3). The following is immediate.

Proposition 1.4. If f : C is holomorphic, then
(1.27)

f
f
and
exist, and are continuous on ,
x
y

and
(1.28)

1 f
f
=
x
i y

on , each side of (1.28) being equal to f 0 on .

24

When (1.27) holds, one says f is of class C 1 and writes f C 1 (). As shown in
Appendix B, if f C 1 (), it is R-differentiable, i.e.,
(1.29)

(1.30)

A=

f
(z),
x

B=

f
(z).
y

(1.31)

for all h1 , h2 R, which holds if and only if

(1.32)

A=

1
B = a,
i

leading back to (1.28). This gives the following converse to Proposition 1.4.
Proposition 1.5. If f : C is C 1 and (1.28) holds, then f is holomorphic.
The equation (1.28) is called the Cauchy-Riemann equation. Here is an alternative
presentation. Write
(1.33)

f (z) = u(z) + iv(z),

u = Re f, v = Im f.

Then (1.28) is equivalent to the system of equations

(1.34)

u
v
=
,
x
y

v
u
= .
x
y

To pursue this a little further, we change perspective, and regard f as a map from an
open subset of R2 into R2 . We represent an element of R2 as a column vector. Objects
on C and on R2 correspond as follows.
On C
z = x + iy
(1.35)

f = u + iv
h = h1 + ih2

On R2

x
z=
y

u
f=
v

h1
h=
h2

25

As discussed in Appendix B, a map f : R2 is differentiable at z if and only if

there exists a 2 2 matrix L such that
(1.36)

f (z + h) = f (z) + Lh + R(z, h),

R(z, h) = o(|h|).

(1.37)

Df (z) =

u/x
v/x

u/y
v/y

The Cauchy-Riemann equations specify that

u
v

, =
(1.38)
Df =
, =
.

x
x
Now the map z 7 iz is a linear transformation on C R2 , whose 2 2 matrix representation is given by

0 1
(1.39)
J=
.
1 0

Note that, if L =
, then

(1.40)
JL =
, LJ =
,

so JL = LJ if and only if = and = . (When JL = LJ, we say J and L commute.)
When L = Df (z), this gives (1.38), proving the following.
Proposition 1.6. If f C 1 (), then f is holomorphic if and only if, for each z ,
(1.41)

Df (z) and J commute.

In the calculus of functions of a real variable, the interaction of derivatives and integrals,
via the fundamental theorem of calculus, plays a central role. We recall the statement.
Theorem 1.7. If f C 1 ([a, b]), then
Z
(1.42)

Furthermore, if g C([a, b]), then, for a < t < b,

(1.43)

d
dt

g(s) ds = g(t).
a

26

In the study of holomorphic functions on an open set C, the partner of d/dz is the
integral over a curve, which we now discuss.
A C 1 curve (or path) in is a C 1 map
: [a, b] ,
where [a, b] = {t R : a t b}. If f : C is continuous, we define
Z
(1.44)

f (z) dz =

f ((t)) 0 (t) dt,

the right side being the standard integral of a continuous function, as studied in beginning
calculus (except that here the integrand is complex valued). More generally, if f, g : C
are continuous and = 1 + i2 , with j real valued, we set
Z
(1.44A)

f (z) dx + g(z) dy =
a

Then (1.44) is the special case g = if (with dz = dx + i dy).

The following result is a counterpart to (1.42).
Proposition 1.8. If f is holomorphic on , and : [a, b] C is a C 1 path, then
Z
(1.45)
f 0 (z) dz = f ((b)) f ((a)).

The proof will use the following chain rule.

Proposition 1.9. If f : C is holomorphic and : [a, b] is C 1 , then, for
a < t < b,
d
f ((t)) = f 0 ((t)) 0 (t).
dt

(1.46)

The proof of Proposition 1.9 is essentially the same as that of Proposition 1.2. To
address Proposition 1.8, we have
Z

Z
0

f (z) dz =
a

(1.47)

f 0 ((t)) 0 (t) dt

d
f ((t)) dt
a dt
= f ((b)) f ((a)),
=

27

the second identity by (1.46) and the third by (1.42). This gives Proposition 1.8.
The second half of Theorem 1.7 involves producing an antiderivative of a given function
g. In the complex context, we have the following.
Definition. A holomorphic function g : C is said to have an antiderivative f on
provided f : C is holomorphic and f 0 = g.
Calculations done above show that g has an antiderivative f in the following cases:

(1.48)

g(z) = z k ,
g(z) = ez ,

1
z k+1 , k 6= 1,
k+1
f (z) = ez .
f (z) =

A function g holomorphic on an open set might not have an antiderivative f on all

of . In cases where it does, Proposition 1.8 implies
Z
(1.49)

g(z) dz = 0

for any closed path in , i.e., any C 1 path : [a, b] such that (a) = (b). In 3,
we will see that if is the unit circle centered at the origin,
Z
(1.50)

1
dz = 2i,
z

so 1/z, which is holomorphic on C \ 0, does not have an antiderivative on C \ 0. In 4, we

will construct log z as an antiderivative of 1/z on the smaller domain C \ (, 0].
We next show that each holomorphic function g : C has an antiderivative for a
significant class of open sets C, namely sets with the following property.

(1.51)

There exists a + ib such that whenever x + iy ,

the vertical line from a + ib to a + iy and the horizontal line
from a + iy to x + iy belong to .

(Here a, b, x, y R.) See Fig. 1.1.

Proposition 1.10. If C is an open set satisfying (1.51) and g : C is holomorphic, then there exists a holomorphic f : C such that f 0 = g.
Proof. Take a + ib as in (1.51), and set, for z = x + iy ,
Z
(1.52)

f (z) = i

g(a + is) ds +
b

a

28

Theorem 1.7 readily gives

f
(z) = g(z).
x

(1.53)
We also have
(1.54)

f
(z) = ig(a + iy) +
y

Z
a

g
(t + iy) dt,
y

and applying the Cauchy-Riemann equation g/y = ig/x gives

(1.55)

Z x
g
1 f
= g(a + iy) +
(t + iy) dt
i y
a t
= g(a + iy) + [g(x + iy) g(a + iy)]
= g(z).

Comparing (1.54) and (1.55), we have the Cauchy-Riemann equations for f , and Proposition 1.10 follows.
Examples of open sets satisfying (1.51) include disks and rectangles, while C \ 0 does
not satisfy (1.51), as one can see by taking a + ib = 1, x + iy = 1.

Exercises
1. Let f, g C 1 (), not necessarily holomorphic. Show that

(1.56)

x

y

on , where fx = f /x, etc.

2. In the setting of Exercise 1, show that, on {z : g(z) 6= 0},
(1.57)
Derive formulas for

1
gx (z)
=
,
x g(z)
g(z)2

1
gy (z)
=
.
y g(z)
g(z)2

f (z)
f (z)
and
.
x g(z)
y g(z)

29

3. In (a)(d), compute f /x and f /y. Determine whether f is holomorphic (and on

what domain). If it is holomorphic, specify f 0 (z).

(c)

z+1
,
z2 + 1
z+1
f (z) = 2
,
z +1
f (z) = e1/z ,

(d)

f (z) = e|z| .

(a)

f (z) =

(b)

4. Find the antiderivative of each of the following functions.

1
,
(z + 3)2

(a)

f (z) =

(b)

f (z) = zez ,

(c)

f (z) = z 2 + ez ,

(t) = t + it2 .
Compute

f (z) dz in the following cases.

(a)

f (z) = z,

(b)

f (z) = z,

(c)

f (z) =

1
,
(z + 5)2
f (z) = ez .

(d)
6. Do Exercise 5 with

(t) = t4 + it2 .

R
7. Recall the definition (1.44) for f (z) dz when f C() and : [a, b] is a C 1
curve. Suppose s : [a, b] [, ] is C 1 , with C 1 inverse, such that s(a) = , s(b) = . Set
(s(t)) = (t). Show that
Z
Z
f (z) dz =

f () d,

30

so path integrals are invariant under change of parametrization.

In the following exercises, let
h f (z) =

1
(f (z + h) f (z)).
h

8. Show that h z 1 z 2 uniformly on {z C : |z| > }, for each > 0.

Hint. Use (1.10).
9. Let C be open and assume K is compact. Assume f, g C() and
h f (z) f 0 (z), h g(z) g 0 (z), uniformly on K.
Show that

h (f (z)g(z)) f 0 (z)g(z) + f (z)g 0 (z), uniformly on K.

Hint. Write
h (f g)(z) = h f (z) g(z + h) + f (z)h g(z).
10. Show that, for each > 0, A < ,
h z n nz (n+1)
uniformly on {z C : < |z| A}.
Hint. Use induction.

31

2. Holomorphic functions defined by power series

A power series has the form
(2.1)

f (z) =

an (z z0 )n .

n=0

Recall from 0 that to such a series there is associated a radius of convergence R [0, ],
with the property that the series converges absolutely whenever |z z0 | < R (if R > 0),
and diverges whenever |z z0 | > R (if R < ). We begin this section by identifying R as
follows:
1
(2.2)
= lim sup |an |1/n .
R
n
This is established in the following result, which reviews and complements Propositions
0.20.3.
Proposition 2.1. The series (2.1) converges whenever |z z0 | < R and diverges whenever
|z z0 | > R, where R is given by (2.2). If R > 0, the series converges uniformly on
{z : |z z0 | R0 }, for each R0 < R. Thus, when R > 0, the series (2.1) defines a
continuous function
(2.3)

f : DR (z0 ) C,

where
(2.4)

Proof. If R0 < R, then there exists N Z+ such that

1
n N = |an |1/n < 0 = |an |(R0 )n < 1.
R
Thus
z z n

0
(2.5)
|z z0 | < R0 < R = |an (z z0 )n |
,
0
R
for n N , so (2.1) is dominated by a convergent geometrical series in DR0 (z0 ).
For the converse, we argue as follows. Suppose R00 > R, so infinitely many |an |1/n
1/R00 , hence infinitely many |an |(R00 )n 1. Then
z z n

0
00
n
|z z0 | R > R = infinitely many |an (z z0 ) |
1,
00
R
forcing divergence for |z z0 | > R.
The assertions about uniform convergence and continuity follow as in Proposition 0.3.
The following result, which extends Proposition 0.4 from the real to the complex domain,
is central to the study of holomorphic functions. A converse will be established in 5, as a
consequence of the Cauchy integral formula.

32

Proposition 2.2. If R > 0, the function defined by (2.1) is holomorphic on DR (z0 ), with
derivative given by
0

(2.6)

f (z) =

nan (z z0 )n1 .

n=1

Proof. Absolute convergence of (2.6) on DR (z0 ) follows as in the proof of Proposition 0.4.
Alternatively (cf. Exercise 3 below), we have
(2.7)

lim n1/n = 1 = lim sup |nan |1/n = lim sup |an |1/n ,

so the power series on the right side of (2.6) converges locally uniformly on DR (z0 ), defining
a continuous function g : DR (z0 ) C. It remains to show that f 0 (z) = g(z).
To see this, consider
(2.8)

fk (z) =

k
X

an (z z0 ) ,

gk (z) =

n=0

k
X

nan (z z0 )n1 .

n=1

We have fk f and gk g locally uniformly on DR (z0 ). By (1.12) we have fk0 (z) = gk (z).
Hence it follows from Proposition 1.8 that, for z DR (z0 ),
Z
(2.9)
fk (z) = a0 + gk () d,
z

for any path z : [a, b] DR (z0 ) such that z (a) = z0 and z (b) = z. Making use of the
locally uniform convergence, we can pass to the limit in (2.9), to get
Z
(2.10)
f (z) = a0 + g() d.
z

Taking z to approach z horizontally, we have (with z = x + iy, z0 = x0 + iy0 )

Z y
Z x
f (z) = a0 +
g(x0 + it) i dt +
g(t + iy) dt,
y0

x0

and hence
f
(z) = g(z),
x

(2.11)

Z x
Z
f (z) = a0 +
g(t + iy0 ) dt +
x0

y0

33

and hence
f
(z) = ig(z).
y

(2.12)

Thus f C 1 (DR (z0 )) and it satisfies the Cauchy-Riemann equation, so f is holomorphic

and f 0 (z) = g(z), as asserted.
Remark. For a proof of Proposition 2.2 making a direct analysis of the difference quotient
[f (z) f (w)]/(z w), see [Ahl].
It is useful to note that we can multiply power series with radius of convergence R > 0.
In fact, there is the following more general result on products of absolutely convergent
series.
Proposition 2.3. Given absolutely convergent series
(2.13)

A=

n ,

B=

n=0

n ,

n=0

(2.14)

AB =

n ,

n =

n=0

Proof. Take Ak =

Pk

n=0 n , Bk =

(2.15)

n
X

j nj .

j=0

Pk
n=0

n . Then

Ak Bk =

k
X

n + Rk

n=0

with
(2.16)

Rk =

(k) = {(m, n) Z+ Z+ : m, n k, m + n > k}.

m n ,

(m,n)(k)

Hence
|Rk |
(2.17)

mk/2 k/2nk

|n | + B

nk/2

|m | |n | +
X

|m | |n |

k/2mk nk

|m |,

mk/2

where
(2.18)

A=

X
n=0

|n | < ,

B=

|n | < .

n=0

It follows thatPRk 0 as k . Thus the left side of (2.15) converges to AB and the

right side to n=0 n . The absolute convergence of (2.14) follows by applying the same
argument with n replaced by |n | and n replaced by |n |.

34

Corollary 2.4. Suppose the following power series converge for |z| < R:

X
X
(2.19)
f (z) =
an z n , g(z) =
bn z n .
n=0

n=0

Then, for |z| < R,

(2.20)

f (z)g(z) =

cn z ,

cn =

n=0

n
X

aj bnj .

j=0

The following result, which is related to Proposition 2.3, has a similar proof.
P
P
Proposition 2.5. If ajk C and j,k |ajk | < , then j ajk is absolutely convergent
P
for each k, k ajk is absolutely convergent for each j, and
XX
X
X X
(2.21)
ajk =
ajk =
ajk .
j

j,k

Proof. Clearly the hypothesis implies j |ajk | < for each k and
j. It also implies that there exists B < such that
SN =

N X
N
X

|ajk | B,

P
k

N.

j=0 k=0

Now SN is bounded and monotone, so there exists a limit, SN % A < as N % . It

follows that, for each > 0, there exists N Z+ such that
X
|ajk | < , C(N ) = {(j, k) Z+ Z+ : j > N or k > N }.
(j,k)C(N )

Now, whenever M, K N ,
M X
K
N X
N
X

ajk
ajk

j=0 k=0

so

j=0 k=0

X
(j,k)C(N )

M X

N X
N
X

ajk
ajk

j=0 k=0

and hence

j=0 k=0

j=0 k=0

|ajk |,

(j,k)C(N )

N X
N
X

ajk
ajk

j=0 k=0

|ajk |,

|ajk |.

(j,k)C(N )

We have a similar result with the roles of j and k reversed, and clearly the two finite sums
agree. It follows that
X

X
X

ajk
ajk < 2, > 0,

j=0 k=0

k=0 j=0

yielding (3.23).
Using Proposition 2.5, we demonstrate the following.

35

Proposition 2.6. If (2.1) has a radius of convergence R > 0, and z1 DR (z0 ), then f (z)
has a convergent power series about z1 :
(2.22)

f (z) =

bk (z z1 )k ,

for |z z1 | < R |z1 z0 |.

k=0

The proof of Proposition 2.6 will not use Proposition 2.2, and we can use this result
to obtain a second proof of Proposition 2.2. Shrawan Kumar showed the author this
argument.

Proof of Proposition 2.6. There is no loss in generality in taking z0 = 0, which we will do

here, for notational simplicity. Setting fz1 () = f (z1 + ), we have from (2.1)
fz1 () =
(2.23)
=

an ( + z1 )n

n=0
X
n
X
n=0 k=0

n k nk
an
z1 ,
k

the second identity by the binomial formula (cf. (2.34) below). Now,
(2.24)

X
n
X
n=0 k=0

X
n
k
nk
|an |
|| |z1 |
=
|an |(|| + |z1 |)n < ,
k
n=0

provided || + |z1 | < R, which is the hypothesis in (2.22) (with z0 = 0). Hence Proposition
2.5 gives
(2.25)

fz1 () =

X
k=0 n=k

n nk k
an
z
.
k 1

(2.26)

bk =

X
n=k

n nk
an
z
.
k 1

This proves Proposition 2.6. Note in particular that

(2.27)

b1 =

X
n=1

nan z1n1 .

36

Second proof of Proposition 2.2. The result (2.22) implies f is complex differentiable
at each z1 DR (z0 ), and the computation (2.27) translates to (2.6), with z = z1 .
Remark. The result of Proposition 2.6 is a special case of a general result on representing
a holomorphic function by a convergent power series, which will be established in 5.

Exercises
1. Determine the radius of convergence R for each of the following series. If 0 < R < ,
examine when convergence holds at points on |z| = R.
(a)

zn

n=0

(b)

X
zn
n
n=1

(c)

X
zn
n2
n=1

(d)

X
zn
n!
n=0

(e)

X
zn
2n
n=0

(f)

X
z2
2n
n=0

2. Show that if the power series (2.1) has radius of convergence R > 0, then f 00 , f 000 , . . .
are holomorphic on DR (z0 ) and
(2.28)

f (n) (z0 ) = n! an .

Here we set f (n) (z) = f 0 (z) for n = 1, and inductively f (n+1) (z) = (d/dz)f (n) (z).

37

(1 + a)n 1 + na.

(2.29)

lim sup n1/n 1,

(2.30)

and hence
lim n1/n = 1,

(2.31)

a result used in (2.7).

Hint. To get (2.30), deduce from (2.29) that n1/n (1 + a)/a1/n . Then show that, for
each a > 0,
lim a1/n = 1.

(2.32)

For another proof of (2.31), see Exercise 4 of 4.

4. The following is a version of the binomial formula. If a C, n N,
(2.33)

n
X
n k
(1 + a) =
a ,
k
n

k=0

n
n!
.
=
k!(n k)!
k

Another version is
(2.34)

n
X
n k nk
(z + w) =
z w
.
k
n

k=0

Verify this identity and show that (2.33) implies (2.29) when a > 0.
Hint. To verify (2.34), expand
(2.35)

(z + w)n = (z + w) (z + w)

as a sum of monomials and count the number of terms equal to z k wnk . Use the fact that

n
(2.36)
= number of combinations of n objects, taken k at a time.
k
5. As a special case of Exercise 2, note that, given a polynomial
(2.37)

p(z) = an z n + + a1 z + a0 ,

38

we have
p(k) (0) = k! ak ,

(2.38)

0 k n.

Apply this to
pn (z) = (1 + z)n .

(2.39)
(k)

Compute pn (z), using (1.5), then compute p(k) (0), and use this to give another proof of
(2.33), i.e.,
(2.40)

n
X
n k
pn (z) =
z ,
k
k=0

n!
n
.
=
k!(n k)!
k

39

3. Exponential and trigonometric functions: Eulers formula

Recall from 0 and 1 that we define the exponential function by its power series:

(3.1)

X zj
z2
zj
e =1+z+
+ +
+ =
.
2!
j!
j!
j=0
z

By the ratio test this converges for all z, to a continuous function on C. Furthermore, the
exponential function is holomorphic on C. This function satisfies
d z
e = ez ,
dz

(3.2)

e0 = 1.

One derivation of this was given in 1. Alternatively, (3.2) can be established by differentiating term by term the series (3.1) to get (by Proposition 2.2)

(3.3)

X 1
d z X
1
e =
z j1 =
zk
dz
(j 1)!
k!
j=1
k=0

X
k=0

1 k
z = ez .
k!

The property (3.2) uniquely characterizes ez . It implies

(3.4)

dj z
e = ez ,
j
dz

j = 1, 2, 3, . . . .

By (2.21), any function f (z) that is the sum of a convergent power series about z = 0 has
the form
(3.5)

f (z) =

X
f (j) (0)
j=0

j!

zj ,

which for a function satisfying (3.2) and (3.4) leads to (3.1). A simple extension of (3.2) is
(3.6)

d az
e = a eaz .
dz

Note how this also extends (0.51).

As shown in (0.60), the exponential function satisfies the fundamental identity
(3.7)

ez ew = ez+w ,

z, w C.

40

For an alternative proof, we can expand the left side of (3.7) into a double series:
ez ew =

(3.8)

X
X
z j X wk
z j wk
=
.
j!
k!
j!k!
j=0
k=0

j,k=0

(3.9)

z+w

X
(z + w)n
=
,
n!
n=0

using the binomial formula (cf. (2.34))

(3.10)

n
X
n j nj
(z + w) =
z w
,
j
j=0
n

n
n!
.
=
j!(n j)!
j

Setting k = n j, we have
(3.11)

z+w

n=0 j+k=n;j,k0

X
z j wk
1 n! j k
z w =
.
n! j!k!
j!k!
j,k=0

See (2.14) for the last identity. Comparing (3.8) and (3.11) again gives the identity (3.7).
We next record some properties of exp(t) = et for real t. The power series (3.1) clearly
gives et > 0 for t 0. Since et = 1/et , we see that et > 0 for all t R. Since
det /dt = et > 0, the function is monotone increasing in t, and since d2 et /dt2 = et > 0,
this function is convex. Note that
(3.12)

et > 1 + t,

for t > 0.

Hence
(3.13)

lim et = +.

t+

Since et = 1/et ,
(3.14)

lim et = 0.

As a consequence,
(3.15)

exp : R (0, )

is smooth and one-to-one and onto, with positive derivative, so the inverse function theorem
of one-variable calculus applies. There is a smooth inverse
(3.16)

L : (0, ) R.

41

(3.17)

log x = L(x).

See Figures 3.1 and 3.2 for graphs of x = et and t = log x.

Applying d/dt to
L(et ) = t

(3.18)
gives
L0 (et )et = 1,

(3.19)

hence L0 (et ) =

1
,
et

i.e.,
d
1
log x = .
dx
x

(3.20)
Since log 1 = 0, we get

Z
(3.21)

log x =
1

dy
.
y

(3.22)

x, y (0, ),

which can also be deduced from (3.21).

We next show how to extend the logarithm into the complex domain, defining log z for
z C \ R , where R = (, 0], using Proposition 1.10. In fact, the hypothesis (1.51)
holds for = C \ R , with a + ib = 1, so each holomorphic function g on C \ R has a
holomorphic anti-derivative. In particular, 1/z has an anti-derivative, and this yields
(3.22A)

log : C \ R C,

d
1
log z = ,
dz
z

log 1 = 0.

By Proposition 1.8, we have

Z
(3.22B)

log z =
1

d
,

the integral taken along any path in C\R from 1 to z. Comparison with (3.21) shows that
this function restricted to (0, ) coincides with log as defined in (3.17). In 4 we display
log in (3.22A) as the inverse of the exponential function exp(z) = ez on the domain
= {x + iy : x R, y (, )}, making use of some results that will be derived next.
(See also Exercises 1315 at the end of this section.)

42

We move next to a study of ez for purely imaginary z, i.e., of

(t) = eit ,

(3.23)

t R.

This traces out a curve in the complex plane, and we want to understand which curve it
is. Let us set
eit = c(t) + is(t),

(3.24)

with c(t) and s(t) real valued. First we calculate |eit |2 = c(t)2 + s(t)2 . For x, y R,
(3.25)

z = x + iy = z = x iy = zz = x2 + y 2 = |z|2 .

It is elementary that
(3.26)

z, w C = zw = z w = z n = z n ,
and z + w = z + w.

Hence
(3.27)

ez

X
zk
k=0

k!

= ez .

In particular,
(3.28)

t R = |eit |2 = eit eit = 1.

Hence t 7 (t) = eit has image in the unit circle centered at the origin in C. Also
(3.29)

so (t) moves at unit speed on the unit circle. We have

(3.30)

(0) = 1,

0 (0) = i.

Thus, for t between 0 and the circumference of the unit circle, the arc from (0) to (t) is
an arc on the unit circle, pictured in Figure 3.3, of length
Z
(3.31)

`(t) =

| 0 (s)| ds = t.

Standard definitions from trigonometry say that the line segments from 0 to 1 and from
0 to (t) meet at angle whose measurement in radians is equal to the length of the arc of
the unit circle from 1 to (t), i.e., to `(t). The cosine of this angle is defined to be the

43

x-coordinate of (t) and the sine of the angle is defined to be the y-coordinate of (t).
Hence the computation (3.31) gives
(3.32)

c(t) = cos t,

s(t) = sin t.

Thus (3.24) becomes

eit = cos t + i sin t,

(3.33)

an identity known as Eulers formula. The identity

d it
e = ieit ,
dt

(3.34)
applied to (3.33), yields

d
cos t = sin t,
dt

(3.35)

d
sin t = cos t.
dt

We can use (1.3.7) to derive formulas for sin and cos of the sum of two angles. Indeed,
comparing
ei(s+t) = cos(s + t) + i sin(s + t)

(3.36)
with

(3.37)
gives
(3.38)

cos(s + t) = (cos s)(cos t) (sin s)(sin t),

sin(s + t) = (sin s)(cos t) + (cos s)(sin t).

Derivations of the formulas (3.35) for the derivative of cos t and sin t given in first
semester calculus courses typically make use of (3.38) and further limiting arguments,
which we do not need with the approach used here.
The standard definition of the number is half the length of the unit circle. Hence
is the smallest positive number such that (2) = 1. We also have
(3.39)

() = 1,

= i.

(3.40)

1
3
= +
i,
2
2

3 1
+ i.
2
2

44

We now show how to compute an accurate approximation to .

This formula will arise by comparing two ways to compute the length of an arc of a
circle. So consider the length of (t) over 0 t . By (3.29) we know it is equal to .
Suppose 0 < < /2 and parametrize this segment of the circle by
(3.41)

p
(s) = ( 1 s2 , s),

0 s = sin .

Z
(3.42)

`=

| (s)| ds =
0

ds
.
1 s2

Comparing these two length calculations, we have

Z

(3.43)
0

ds
= ,
1 s2

sin = ,

when 0 < < /2. As anotherway to see this, note that the substitution s = sin gives,
by (3.35), ds = cos d, while 1 s2 = cos by (3.28), which implies
cos2 t + sin2 t = 1.

(3.44)
Thus
Z
(3.45)
0

ds

=
1 s2

d = ,
0

again verifying (3.43).

In particular, using sin(/6) = 1/2, from (3.40), we deduce that
(3.46)

=
6

1/2

dx
.
1 x2

One can produce a power series for (1 y)1/2 and substitute y = x2 . (For more on
this, see the exercises at the end of 5.) Integrating the resulting series term by term, one
obtains
(3.47)

an 1 2n+1
=
,
6
2n
+
1
2
n=0

where the numbers an are defined inductively by

(3.48)

a0 = 1,

an+1 =

2n + 1
an .
2n + 2

45

Using a calculator, one can sum this series over 0 n 20 and show that
(3.49)

= 3.141592653589 .

We leave the verification of (3.47)(3.49) as an exercise, once one gets to Exercises 23 of

5.

Exercises
Heres another way to demonstrate the formula (3.35) for the derivatives of sin t and cos t.
1. Suppose you define cos t and sin t so that (t) = (cos t, sin t) is a unit-speed parametrization of the unit circle centered at the origin, satisfying (0) = (1, 0), 0 (0) = (0, 1), (as we
did in (3.32)). Show directly (without using (3.35)) that
0 (t) = ( sin t, cos t),
and hence deduce (3.35). (Hint. Differentiate (t) (t) = 1 to deduce that, for each
t, 0 (t) (t). Meanwhile, |(t)| = | 0 (t)| = 1.)
2. It follows from (3.33) and its companion eit = cos t i sin t that
(3.50)

cos z =

eiz + eiz
,
2

sin z =

eiz eiz
2i

for z = t R. We define cos z and sin z as holomorphic functions on C by these identities.

Show that they yield the series expansions
(3.51)

cos z =

X
(1)k
k=0

(2k)!

2k

z ,

X
(1)k
sin z =
z 2k+1 .
(2k + 1)!
k=0

3. Extend the identities (3.35) and (3.38) to complex arguments. In particular, for z C,
we have
(3.52)

cos(z + 2 ) = sin z,
sin(z +

2)

= cos z,

cos(z + ) = cos z,

cos(z + 2) = cos z,

sin(z + ) = sin z,

sin(z + 2) = sin z.

4. We define
(3.53)

cosh y =

ey + ey
,
2

sinh y =

ey ey
.
2

(3.54)

cos(x + iy) = cos x cosh y i sin x sinh y,

sin(x + iy) = sin x cosh y + i cos x sinh y.

46

5. Define tan z for z 6= (k + 1/2) and cot z for z 6= k, k Z, by

(3.55)

tan z =

sin z
,
cos z

cot z =

cos z
.
sin z

Show that
(3.56)

tan(z + 2 ) = cot z,

tan(z + ) = tan z,

and
(3.57)

d
1
tan z =
= 1 + tan2 z.
dz
cos2 z

6. For each of the following functions g(z), find a holomorphic function f (z) such that
f 0 (z) = g(z).
a) g(z) = z k ez , k Z+ .
b) g(z) = eaz cos bz.
7. Concerning the identities in (3.40), verify algebraically that

3 3
+
i = 1.
2
2

Then use ei/6 = ei/2 ei/3 to deduce the stated identity for (/6) = ei/6 .
8. Let be the unit circle centered at the origin in C, going counterclockwise. Show that
Z

1
dz =
z

ieit
dt = 2i,
eit

as stated in (1.50).
9. One sets
sec z =

1
,
cos z

so (3.57) yields (d/dz) tan z = sec2 z. Where is sec z holomorphic? Show that
d
sec z = sec z tan z.
dz
10. Show that
1 + tan2 z = sec2 z.

47

11. Show that

and

d
(sec z + tan z) = sec z (sec z + tan z),
dz
d
(sec z tan z) = sec z tan2 z + sec3 z
dz
= 2 sec3 z sec z.

(Hint. Use Exercise 10 for the last identity.)

12. The identities (3.53) serve to define cosh y and sinh y for y C, not merely for y R.
Show that
d
d
cosh z = sinh z,
sinh z = cosh z,
dz
dz
and
cosh2 z sinh2 z = 1.
The next exercises present log, defined on C \ R as in (3.22A), as the inverse function to
the exponential function exp(z) = ez , by a string of reasoning different from what we will
use in 4. In particular, here we avoid having to appeal to the inverse function theorem,
Theorem 4.2. Set
(3.58)

= {x + iy : x R, y (, )}.

13. Using ex+iy = ex eiy and the properties of exp : R (0, ) and of (y) = eiy
established in (3.15) and (3.33), show that
(3.59)

14. Show that

(3.60)

log(ez ) = z

z .

Hint. Apply the chain rule to compute g 0 (z) for g(z) = log(ez ). Note that g(0) = 0.
15. Deduce from Exercises 13 and 14 that
log : C \ R
is one-to-one and onto, and is the inverse to exp in (3.59).

48

4. Square roots, logs, and other inverse functions

We recall the Inverse Function Theorem for functions of real variables.
Theorem 4.1. Let Rn be open and let f : Rn be a C 1 map. Take p and
assume Df (p) End(Rn ) is invertible. Then there exists a neighborhood O of p and a
neighborhood U of q = f (p) such that f : O U is one-to-one and onto, the inverse
g = f 1 : U O is C 1 , and, for x O, y = f (x),
(4.1)

Dg(y) = Df (x)1 .

A proof of this is given in Appendix B. This result has the following consequence, which
is the Inverse Function Theorem for holomorphic functions.
Theorem 4.2. Let C be open and let f : C be holomorphic. Take p
and assume f 0 (p) 6= 0. Then there exists a neighborhood O of p and a neighborhood U of
q = f (p) such that f : O U is one-to-one and onto, the inverse g = f 1 : U O is
holomorphic, and, for z O, w = f (z),
(4.2)

g 0 (w) =

1
f 0 (z)

Proof. If we check that g is holomorphic, then (4.2) follows from the chain rule, Proposition
1.2, applied to
g(f (z)) = z.
We know g is C 1 . By Proposition 1.6, g is holomorphic on U if and only if, for each w U ,
Dg(w) commutes with J, given by (1.39). Also Proposition 1.6 implies Df (z) commutes
with J. To finish, we need merely remark that if A is an invertible 2 2 matrix,
(4.3)

AJ = JA A1 J = JA1 .

As a first example, consider the function Sq(z) = z 2 . Note that we can use polar
coordinates, (x, y) = (r cos , r sin ), or equivalently z = rei , obtaining z 2 = r2 e2i . This
shows that Sq maps the right half-plane
(4.4)

H = {z C : Re z > 0}

bijectively onto C \ R (where R = (, 0]). Since Sq0 (z) = 2z vanishes only at z = 0,

we see that we have a holomorphic inverse
(4.5)

Sqrt : C \ R H,

49

given by
(4.6)

r > 0, < < .

We also write
z 1/2 = Sqrt(z).

(4.7)

We can define other non-integral powers of z on C \ R . Before doing so, we take a look
at log, the inverse function to the exponential function, exp(z) = ez . Consider the strip
(4.8)

= {x + iy : x R, < y < }.

Since ex+iy = ex eiy , we see that we have a bijective map

exp : C \ R .

(4.9)

Note that dez /dz = ez is nowhere vanishing, so (4.9) has a holomorphic inverse we denote
log:
log : C \ R .

(4.10)
Note that
(4.11)

log 1 = 0.

Applying (4.2) we have

d z
d
1
e = ez =
log z = .
dz
dz
z
Thus, applying Proposition 1.8, we have
Z z
1
(4.13)
log z =
d,
1
(4.12)

where the integral is taken along any path from 1 to z in C \ R . Comparison with (3.22B)
shows that the function log produced here coincides with the function arising in (3.22A).
(This result also follows from Exercises 1315 of 3.)
Now, given any a C, we can define
(4.14)

z a = Powa (z),

Powa : C \ R C

by
z a = ea log z .

(4.15)
The identity eu+v = eu ev then gives
(4.16)

z a+b = z a z b ,

a, b C, z C \ R .

In particular, for n Z, n 6= 0,
(4.17)

(z 1/n )n = z.

Making use of (4.12) and the chain rule (1.20), we see that
d a
(4.18)
z = a z a1 .
dz
While Theorem 4.1 and Corollary 4.2 are local in nature, the following result can provide
global inverses, in some important cases.

50

Proposition 4.3. Suppose C is convex. Assume f is holomorphic in and there

exists a C such that
Re af 0 > 0 on .
Then f maps one-to-one onto its image f ().
Proof. Consider distinct points z0 , z1 . The convexity implies the line (t) = (1
t)z0 + tz1 is contained in , for 0 t 1. By Proposition 1.8, we have
Z 1

f (z1 ) f (z0 )
(4.19)
a
=
af 0 (1 t)z0 + tz1 dt,
z1 z0
0
which has positive real part and hence is not zero.
As an example, consider the strip
o
n
e = x + iy : < x < , y R .
(4.20)

2
2
0
Take f (z) = sin z, so f (z) = cos z. It follows from (3.54) that
(4.21)

e
Re cos z = cos x cosh y > 0, for z ,

so f maps
(4.22)

sin z = g(eiz ), where g() =

1
1

,
2i

e under z 7 eiz is the right half plane H, given by (4.4). Below we will
and the image of
show that the image of H under g is
(4.23)

e one-to-one onto the set (4.23). The inverse function is

It then follows that sin maps
1
denoted sin :
e
(4.24)
sin1 : C \ {(, 1] [1, )} .
e and it follows that
We have sin2 z C \ [1, ) for z ,
(4.25)

e
z .

Hence, by (4.2), g(z) = sin1 z satisfies

(4.26)

g 0 (z) = (1 z 2 )1/2 ,

(4.27)

sin

z C \ {(, 1] [1, )},

Z

z=

(1 2 )1/2 d,

where the integral is taken along any path from 0 to z in C\{(, 1][1, )}. Compare
this identity with (3.43), which treats the special case of real z (1, 1).
It remains to prove the asserted mapping property of g, given in (4.22). We rephrase
the result for
1
1
(4.28)
h() = g(i) =
+
.
2

51

Proposition 4.4. The function h given by (4.28) maps both the upper half plane U = { :
Im > 0} and the lower half plane U = { : Im < 0}, one-to-one onto the set (4.23).
Proof. Note that h : C \ 0 C, and
(4.29)

= h().

Taking w C, we want to solve h() = w for . This is equivalent to

(4.30)

2 2w + 1 = 0,

with solutions
(4.31)

=w

p
w2 1.

Thus, for each w C, there are two solutions, except for w = 1, with single solutions
h(1) = 1, h(1) = 1. If we examine h(x) for x R \ 0 (see Fig. 4.1), we see that
w (, 1] [1, ) if and only if R. If w belongs to the set (4.23), h() = w has
two solutions, both in C \ R, and by (4.29) they are reciprocals of each other. Now
(4.32)

= 2,

||

so, given C \ R, we have U 1/ U . This proves Proposition 4.4.

Exercises
1. Show that, for |z| < 1,
(4.33)

log(1 + z) =

(1)n1

n=1

Z
log(1 + z) =
0

zn
.
n

1
d,
1+

and plug in the power series for 1/(1 + ).

2. Using Exercise 1 (plus further arguments), show that
(4.34)

X
(1)n1
= log 2.
n
n=1

52

Hint. Using properties of alternating series, show that, for r (0, 1),
(4.34A)

N
X
(1)n1 n
r = log(1 + r) + N (r),
n
n=1

|N (r)|

rN +1
.
N +1

Then let r 1 in (4.34A).

3. Take x (0, ). Show that
lim

log x
= 0.
x

Hint. If x = ey , (log x)/x = yey .

4. Using Exercise 3, show that
lim x1/x = 1.

x+

Note that this contains the result (2.31).

Hint. x1/x = e(log x)/x .
5. Write the Euler identity as
eiw =

1 z 2 + iz,

z = sin w,

(4.35)

sin1 z =

1
log 1 z 2 + iz ,
i

|z| < 1.

Does this extend to C \ {(, 1] [1, )}?

6. Compute the following quantities:
a) i1/2 ,
b) i1/3 ,
c) ii .

e given by (4.20) diffeomorphically onto

7. Show that tan z maps the strip
(4.36)

C \ {(, 1]i [1, )i}.

Hint. Consult Fig. 4.2. To get the last step, it helps to show that
R(z) =

1z
1+z

53

has the following properties:

(a) R : C \ {1} C \ {1}, and R(z) = w z = R(w),
(b) R : R \ {1} R \ {1},
(c) R : (0, ) (1, 1),
and in each case the map is one-to-one and onto.
8. Making use of (3.57), show that on the region (4.36) we have
Z
(4.37)

tan

z=
0

d
,
1 + 2

where tan1 is the inverse of tan in Exercise 7, and the integral is over any path from 0
to z within the region (4.36).
9. Show that

d
log sec z = tan z.
dz

On what domain in C does this hold?

10. Using Exercise 11 of 3, compute
d
log(sec z + tan z),
dz
and find the antiderivatives of
sec z

and sec3 z.

On what domains do the resulting formulas work?

11. Consider the integral
Z
(4.38)

1 + t2 dt,

on an interval [0, x] R.
(a) Evaluate (4.38) using the change of variable t = tan and the results of Exercise 10.
(b) Evaluate (4.38) using the change of variable t = sinh u and the results of Exercise 12
in 3.
12. As a variant of (4.20)(4.23), show that z 7 sin z maps
n
o

2
2

54

one-to-one and onto the upper half plane {z : Im z > 0}.

The next exercises construct the holomorphic inverse to sin on the set
(4.39)

= C \ {(, 1] [1, )},

satisfying (4.27), in a manner that avoids appeal to Theorem 4.2. (Compare the construction of log as an inverse to exp in Exercises 1315 of 3.) We will use the result
that
(4.40)

sin :

e as in (4.20), established above via Proposition 4.4.

with
13. Show that, for as in (4.39),
z z 2 C \ [1, ) 1 z 2 C \ (, 0],
and deduce that the function f (z) = (1 z 2 )1/2 is holomorphic on .
14. Show that the set in (4.39) satisfies the hypotheses of Proposition 1.10, with a + ib =
0. Deduce that the function f (z) = (1 z 2 )1/2 has a holomorphic anti-derivative G on
:
(4.41)

G0 (z) = (1 z 2 )1/2 ,

G : C,

G(0) = 0.

Z
(4.42)

G(z) =

(1 2 )1/2 d,

z ,

the integral taken along any path in from 0 to z.

15. Show that
(4.43)

G(sin z) = z,

e
z .

Hint. Apply the chain rule to f (z) = G(sin z), making use of (4.25), to show that f 0 (z) = 1
e
for all z .
16. Use (4.40) and Exercise 15 to show that
(4.44)

G :

and is the holomorphic inverse to sin in (4.40).

55

17. Expanding on Proposition 4.4, show that the function h, given by (4.28), has the
following properties:
(a) h : C \ 0 C is onto,
(b) h : C \ {0, 1, 1} C \ {1, 1} is two-to-one and onto,
(c) h : R \ 0 R \ (1, 1) is onto, and is two-to-one, except at x = 1.
18. Given a C \ R , set
Ea (z) = az = ez log a ,

z C.

Show that Ea is holomorphic in z and compute Ea0 (z).

19. Take the following path to explicitly finding the real and imaginary parts of a solution
to
z 2 = a + ib,
given a + ib
/ R . Namely, with x = Re z, y = Im z, we have
x2 y 2 = a,
and also

x2 + y 2 = =

hence

r
x=

+a
,
2

2xy = b,
p

a2 + b2 ,

y=

b
.
2x

56

5. The Cauchy integral theorem and the Cauchy integral formula

The Cauchy integral theorem is of fundamental importance in the study of holomorphic
functions on domains in C. Our first proof will derive it from Greens theorem, which we
now state.
Theorem 5.1. If is a bounded region in R2 with piecewise smooth boundary, and f and
g belong to C 1 (), then
(5.1)

ZZ
Z
f
g

dx dy = (f dx + g dy).
x y

This result is proven in many texts on multivariable calculus. We give a proof in

Appendix D.
We will apply Greens theorem to the line integral
Z
(5.2)

Z
f dz =

f (dx + i dy).

Clearly (5.1) applies to complex-valued functions, and if we set g = if, we get

Z
(5.3)

ZZ
f
f
f dz =
i

dx dy.
x
y

Whenever f is holomorphic, the integrand on the right side of (5.3) vanishes, so we have
the following result, known as Cauchys integral theorem:
Theorem 5.2. If f C 1 () is holomorphic, then
Z
(5.4)

f (z) dz = 0.

Until further notice, we assume is a bounded region in C, with smooth boundary.

Using (5.4), we can establish Cauchys integral formula:
Theorem 5.3. If f C 1 () is holomorphic and z0 , then
(5.5)

1
f (z0 ) =
2i

f (z)
dz.
z z0

57

Proof. Note that g(z) = f (z)/(z z0 ) is holomorphic on \ {z0 }. Let Dr be the open disk
of radius r centered at z0 . Pick r so small that Dr . Then (5.4) implies
Z
Z
f (z)
f (z)
(5.6)
dz =
dz.
z z0
z z0

Dr

To evaluate the integral on the right, parametrize the curve Dr by () = z0 +rei . Hence
dz = irei d, so the integral on the right is equal to
Z
(5.7)
0

f (z0 + rei )
irei d = i
rei

f (z0 + rei ) d.

As r 0, this tends in the limit to 2if (z0 ), so (5.5) is established.

Note that, when (5.5) is applied to = Dr , the disk of radius r centered at z0 , the
computation (5.7) yields
(5.8)

1
f (z0 ) =
2

2
0

1
f (z0 + re ) d =
`(Dr )

f (z) ds(z),
Dr

when f is holomorphic and C 1 on Dr , and `(Dr ) = 2r is the length of the circle Dr .

This is a mean value property. We will extend this to harmonic functions in a later section.
Let us rewrite (5.5) as
Z
1
f ()
(5.9)
f (z) =
d,
2i
z

Z
1
f ()
0
(5.10)
f (z) =
d,
2i
( z)2

(5.11)

(n)

n!
(z) =
2i

f ()
d.
( z)n+1

In more detail, with

h f (z) =
(5.9) gives

1
(f (z + h) f (z)),
h

1
h f (z) =
2i

Z
h

1
f () d.
z

58

Now, as h 0, given z
/ ,
h ( z)1 ( z)2 , uniformly on ,
by Exercise 8 of 1, and this gives (5.10). The formula (5.11) follows inductively, using
Z

(n 1)!
1
(n1)
h f
(z) =
h
f () d,
2i
( z)n

and applying Exercise 10 of 1.

Here is one consequence of (5.10)(5.11).
Corollary 5.4. Whenever f is holomorphic on an open set C, we have
f C ().

(5.12)

Suppose f C 1 () is holomorphic, z0 Dr , where Dr is the disk of radius r

centered at z0 , and suppose z Dr . Then Theorem 5.3 implies
Z
1
f ()
(5.13)
f (z) =
d.
2i
( z0 ) (z z0 )

(5.14)

1
1 X z z0 n
=
,
( z0 ) (z z0 )
z0 n=0 z0

valid as long as |z z0 | < | z0 |. Hence, given |z z0 | < r, this series is uniformly

convergent for , and we have
Z
f () z z0 n
1 X
d.
(5.15)
f (z) =
2i n=0
z0 z0

This establishes the following key result.

Theorem 5.5. If f C 1 () is holomorphic, then for z Dr (z0 ) , f (z) has the
convergent power series expansion
(5.16)

f (z) =

an (z z0 )n ,

n=0

with
(5.17)

1
an =
2i

f ()
f (n) (z0 )
d
=
.
( z0 )n+1
n!

Remark. The second identity in (5.17) follows from (5.11). Alternatively, once we have
(5.16), it also follows from (2.28) that an is equal to the last quantity in (5.17).
Next we use the Cauchy integral theorem to produce an integral formula for the inverse
of a holomorphic map.

59

Proposition 5.6. Suppose f is holomorphic and one-to-one on a neighborhood of , a

closed bounded domain in C. Set g = f 1 : f () . Then
Z
1
zf 0 (z)
(5.18)
g(w) =
dz, w f ().
2i
f (z) w

Proof. Set = g(w), so h(z) = f (z) w has one zero in , at z = , and h0 () 6= 0.

(Cf. Exercise 8 below.) Then the right side of (5.18) is equal to
Z
Z
1
h0 (z)
1
0 (z)
1
z
z
dz =
+
dz = ,
(5.19)
2i
h(z)
2i
z
(z)

where we set h(z) = (z )(z) with holomorphic and nonvanishing on a neighborhood

of .
Having discussed fundamental consequences of Cauchys theorem, we return to the
theorem itself, and give three more proofs. We begin with the following result, closely
related though not quite identical, to Theorem 5.2. We give a proof using not Greens
theorem but simply the chain rule, the fundamental theorem of calculus, and the equality
of mixed partial derivatives for C 2 functions of two real variables.
Proposition 5.7. Let f C 1 () be holomorphic. Let s be a smooth family of smooth
(class C 2 ) closed curves in . Then
Z
(5.20)
f (z) dz = A
s

is independent of s.
To set things up, say s (t) = (s, t) is C 2 for a s b, t R, and periodic of period 1
in t. Denote the left side of (5.20) by (s):
Z 1

(5.21)
(s) =
f (s, t) t (s, t) dt.
0

Hence
Z
(5.22)

(s) =
0

We compare this with

Z 1

f (s, t) s (s, t) dt
0 t
(5.23)
Z 1
0

=
f (s, t) t (s, t) s (s, t) + f (s, t) t s (s, t) dt.
0

60

Using the identity s t (s, t) = t s (s, t) and the identity

(5.24)

f 0 ()s t = f 0 ()t s ,

we see that the right sides of (5.22) and (5.23) are equal. But the fundamental theorem of
calculus implies the left side of (5.23) is equal to

(5.25)
f (s, 1) s (s, 1) f (s, 0) s (s, 0) = 0.
Thus 0 (s) = 0 for all s, and the proposition is proven.
For a variant of Proposition 5.7, see Exercise 14.
Our third proof of Cauchys theorem establishes a result slightly weaker than Theorem
5.1, namely the following.
Proposition 5.8. With as in Theorem 5.1, assume O, open in C, and f is
holomorphic on O. Then (5.4) holds.
The proof will not use Greens thorem. Instead, it is based on Propositions 1.8 and
1.10.
By Proposition 1.8, if f had a holomorphic antiderivative on O, then wed have
R
f
(z)
dz = 0 for each closed path in O. Since is a union of such closed paths,

this would give (5.4). As we have seen, f might not have an antiderivative on O, though
Proposition 1.10 does give conditions guaranteeing the existence of an antiderivative. The
next strategy is to chop some neighborhood of in O into sets to which Proposition 1.10
applies.
To carry this out, tile the plane C with closed squares Rjk of equal size, with sides
parallel to the coordinate axes, and check whether the following property holds:
(5.26)

If Rjk intersects , then Rjk O.

See Fig. 5.1 for an example of part of such a tiling. If (5.26) fails, produce a new tiling by
dividing each square into four equal subsquares, and check (5.26) again. Eventually, for
example when the squares have diameters less than dist(, O), which is positive, (5.26)
must hold.
If Rjk 6= , denote this intersection by jk . We have = j,k jk , and furthermore
Z
X Z
(5.27)
f (z) dz =
f (z) dz,
j,k

jk

the integrals over those parts of jk not in cancelling out. Now Proposition 1.10
implies f has a holomorphic antiderivative on each Rjk O, and then Proposition 1.8
implies
Z
(5.28)
f (z) dz = 0,
jk

so (5.4) follows.
Finally, we relax the hypotheses on f , for a certain class of domains . To specify the
class, we say an open set O C is star shaped if there exists p O such that
(5.28A)

0 < a < 1, p + z O = p + az O.

61

Theorem 5.9. Let be a bounded domain with piecewise smooth boundary. Assume can
be partitioned into a finite number of piecewise smoothly bounded domains j , 1 j K,
such that each j is star shaped. Assume f C() and that f is holomorphic on . Then
(5.4) holds.
Proof. Since
Z
(5.28B)

f (z) dz =

K Z
X

f (z) dz,

j=1
j

it suffices to prove the result when itself is star shaped, so (5.28A) holds with O = ,
p . Given f C(), holomorphic on , define
(5.28C)

fa : C,

fa (p + z) = f (p + az),

0 < a < 1.

Then Proposition 5.8 (or Theorem 5.2) implies

Z
fa (z) dz = 0 for each a < 1.
(5.28D)

On the other hand, since f is continuous on , fa f uniformly on (and in particular

on ) as a % 1, so (5.4) follows from (5.28D) in the limit as a % 1.

Exercises
1. Using (5.10), show that if fk are holomorphic on C and fk f locally uniformly,
then f is holomorphic and fk f locally uniformly.
2. Show that, for |z| < 1, C,
(5.29)

(1 + z) =

an ()z n ,

n=0

(5.30)

an () =

( 1) ( n + 1)
.
n!

Hint. Use (4.18) to compute f (n) (0) when f (z) = (1 + z) .

3. Deduce from Exercise 2 that, for |z| < 1,
(5.31)

(1 z ) =

(1)n an ()z 2n ,

n=0

62

with an () as above. Take = 1/2 and verify that (3.46) yields (3.47).
4. Suppose f is holomorphic on a disk centered at 0 and satisfies
f 0 (z) = af (z),
for some a C. Prove that f (z) = Keaz for some K C.
Hint. Find the coefficients in the power series for f about 0. Alternative. Apply d/dz to
eaz f (z).
5. Suppose f : C C is a function that is not identically zero. Assume that f is
complex-differentiable at the origin, with f 0 (0) = a. Assume that
f (z + w) = f (z)f (w)
for all z, w C. Prove that f (z) = eaz .
Hint. Begin by showing that f is complex-differentiable on all of C.
6. Suppose f : C is holomorphic, p , and f (p) = 0. Show that g(z) = f (z)/(z p),
defined at p as g(p) = f 0 (p), is holomorphic. More generally, if f (p) = = f (k1) (p) = 0
and f (k) (p) 6= 0, show that f (z) = (z p)k g(z) with g holomorphic on and g(p) 6= 0.
Hint. Consider the power series of f about p.
7. For f as in Exercise 6, show that on some neighborhood of p we can write f (z) =
[(z p)h(z)]k , for some nonvanishing holomorphic function h.
8. Suppose f : C is holomorphic and one-to-one. Show that f 0 (p) 6= 0 for all p .
Hint. If f 0 (p) = 0, then apply Exercise 7 (to f (z) f (p)), with some k 2. Apply
Theorem 4.2 to the function G(z) = (z p)h(z).
Reconsider this problem when you get to 11, and again in 17.
9. Assume f : C is holomorphic, Dr (z0 ) , and |f (z)| M for z Dr (z0 ). Show
that
(5.32)

|f (n) (z0 )|
M
n.
n!
r

Hint. Use (5.11), with replaced by Dr (z0 ).

These inequalities are known as Cauchys inequalities.
A connected open set C is said to be simply connected if each smooth closed curve
in is part of a smooth family of closed curves s , 0 s 1, in , such that 1 = and
0 (t) has a single point as image.
10. Show that if C is open and simply connected, is a smooth closed curve in ,

63

and f is holomorphic on , then

f (z) dz = 0.

11. Take = {z C : 0 < |z| < 2}, and let (t) = eit , 0 t 2. Calculate
deduce that is not simply connnected.

dz/z and

12. Show that if C is open and convex, then it is simply connected.

13. Show that if C is open and simply connected and f is holomorphic on , then f
has a holomorphic antiderivative on .
14. Modify the proof of Proposition 5.7 to establish the following.
Proposition 5.7A. Let C be open and connected. Take p, q and let s be a
smooth family of curves s : [0, 1] such that s (0) p and s (1) q. Let f be
holomorphic on . Then
Z
f (z) dz = A
s

is independent of s.
For more on this, see Exercise 8 of 7.

64

6. The maximum principle, Liouvilles theorem, and the fundamental theorem

of algebra
Here we will apply results of 5 and derive some useful consequences. We start with the
mean value property (5.8), i.e.,
1
f (z0 ) =
2

(6.1)

f (z0 + rei ) d,

valid whenever
(6.2)

Dr (z0 ) = {z C : |z z0 | r} ,

provided f is holomorphic on an open set C. Note that, in such a case,

Z

ZZ

f (z) dx dy =
(6.3)

Dr (z0 )

f (z0 + sei )s ds d

= r2 f (z0 ),
or
(6.4)

1
f (z0 ) =
Ar

ZZ
f (z) dx dy,

Dr (z0 )

where Ar = r2 is the area of the disk Dr (z0 ). This is another form of the mean value
property. We use it to prove the following result, known as the maximum principle for
holomorphic functions.
Proposition 6.1. Let C be a connected, open set. If f is holomorphic on , then,
given z0 ,
(6.5)

z

(6.6)

sup |f (z)| = sup |f (z)|.

z

Proof. In the latter context, |f | must assume a maximum at some point in (cf. Proposition A.14). Hence it suffices to prove (6.5).

65

Thus, assume there exists z0 such that the hypotheses of (6.5) hold. Set
(6.7)

O = { : f () = f (z0 )}.

We have z0 O. Continuity of f on implies O is a closed subset of . Now, if 0 O,

there is a disk of radius , D (0 ) , and, parallel to (6.9),
ZZ
1
(6.8)
f (0 ) =
f (z) dx dy.
A
D (0 )

(6.9)

f (0 ) = f (z),

z D (0 ).

Hence O is an open subset of , as well as a nonempty closed subset. As explained in

Appendix A, the hypothesis that is connected then implies O = . This completes the
proof.
One useful consequence of Proposition 6.1 is the following result, known as the Schwarz
lemma.
Proposition 6.2. Suppose f is holomorphic on the unit disk D1 (0). Assume |f (z)| 1
for |z| < 1, and f (0) = 0. Then
(6.10)

|f (z)| |z|.

Furthermore, equality holds in (6.10), for some z D1 (0) \ 0, if and only if f (z) = cz for
some constant c of absolute value 1.
Proof. The hypotheses imply that g(z) = f (z)/z is holomorphic on D1 (0) (cf. 5, Exercise
6), and that |g(z)| 1/a on the circle {z : |z| = a}, for each a (0, 1). Hence the
maximum principle implies |g(z)| 1/a on Da (0). Letting a % 1, we obtain |g(z)| 1 on
D1 (a), which implies (6.10).
If |f (z0 )| = |z0 | at some point in D1 (0), then |g(z0 )| = 1, so Proposition 6.1 implies
g c, hence f (z) cz.
The next result is known as Liouvilles theorem.
Proposition 6.3. If f : C C is holomorphic and bounded, then f is constant.
Proof. Given z C, we have by (5.10), for each R (0, ),
Z
1
f ()
0
(6.11)
f (z) =
d,
2i
( z)2
DR (z)

where
(6.12)

DR (z) = { C : | z| = R}.

66

1
f (z) =
2i

(6.13)

1
=
2R

f (z + Reit )
iReit dt
R2 e2it

hence, if |f (z)| M for all z C,

|f 0 (z)|

(6.14)

M
.
R

Compare the case n = 1 of (5.32). Since (6.13) holds for all R < , we obtain
f 0 (z) = 0,

(6.15)

z C,

which implies f is constant.

Second proof of Proposition 6.3. With f (0) = a, set
g(z) =

f (z) a
z
0
f (0)

for z 6= 0,
for z = 0.

Then g : C C is holomorphic (cf. 5, Exercise 6). The hypothesis |f (z)| M for all z
implies
M + |a|
|g(z)|
for |z| = R,
R
so the maximum principle implies |g(z)| (M + |a|)/R on DR (0), and letting R
gives g 0, hence f a.
We are now in a position to prove the following result, known as the fundamental
theorem of algebra.
Theorem 6.4. If p(z) = an z n + an1 z n1 + + a1 z + a0 is a polynomial of degree n 1
(an 6= 0), then p(z) must vanish somewhere in C.
Proof. Consider
(6.16)

f (z) =

1
.
p(z)

If p(z) does not vanish anywhere on C, then f (z) is holomorphic on all of C. On the other
hand, when z 6= 0,
(6.17)

f (z) =

1
1
,
n
1
z an + an1 z + + a0 z n

67

so
(6.18)

|f (z)| 0, as |z| .

Thus f is bounded on C, if p(z) has no roots. By Proposition 6.3, f (z) must be constant,
which is impossible, so p(z) must have a complex root.
Alternatively, having (6.18), we can apply the maximum principle. Applied to f (z)
on DR (0), it gives |f (z)| sup||=R |f ()| for |z| R, and (6.18) then forces f to be
identically 0, which is impossible.
See Appendix F for an elementary proof of the fundamental theorem of algebra, i.e.,
a proof that does not make use of results from integral calculus.

Exercises
1. Establish the following improvement of Liouvilles theorem.
Proposition. Assume f : C C is holomorphic, and that there exist w0 C and a > 0
such that
|f (z) w0 | a, z C.
Then f is constant.
Hint. Consider

1
.
f (z) w0

g(z) =

2. Let A C be an open annulus, with two boundary components, 0 and 1 . Assume

f C(A) is holomorphic in A. Show that one cannot have
(6.19)

Hint. Assume (6.21) holds. Then K = {z A : Re f (z) = 0} is a nonempty compact

subset of A (disjoint from A), and J = {f (z) : z K} is a nonempty compact subset
of the imaginary axis. Pick ib J so that b is maximal. Show that, for sufficiently small
> 0,
1
g (z) =
,
i(b + ) f (z)
which is holomorphic on A, would have to have an interior maximum, which is not allowed.
3. Show that the following calculation leads to another proof of the mean value property
for f holomorphic on C, when DR (p) .
(6.20)

1
(r) =
2

Z
0

f (p + rei ) d

68

satisfies
1
(r) =
2

(6.21)

f 0 (p + rei )ei d

1
=
2ir

Z
0

d
f (p + rei ) d.
d

4. Let = {z C : 0 < Re z < 1}. Assume f is bounded and continuous on and

holomorphic on . Show that
sup |f | = sup |f |.

Hint. For > 0, consider f (z) = f (z)ez .

Relax the hypothesis that f is bounded.
For Exercises 58, suppose we have a polynomial p(z), of the form
(6.22)

p(z) = z n + an1 z n1 + + a1 z + a0 .

5. Show that there exist rk C, 1 k n, such that

(6.23)

p(z) = (z r1 ) (z rn ).

6. Show that
(6.24)

p0 (z)
1
1
=
+ +
.
p(z)
z r1
z rn

7. Suppose each root rk of p(z) belongs to the right half-plane H = {z : Re z > 0}. Show
that
(6.25)

Re z < 0 = Re

p0 (z)
p0 (z)
< 0 =
6= 0.
p(z)
p(z)

8. Show that the set of zeros of p0 (z) is contained in the convex hull of the set of zeros of
p(z).
Hint. Given a closed set S Rn , the convex hull of S is the intersection of all the
half-spaces containing S.
9. Suppose f : C C is holomorphic and satisfies an estimate
|f (z)| C(1 + |z|)n1 ,
for some n Z+ . Show that f (z) is a polynomial in z of degree n 1.
Hint. Apply (5.32) with = DR (z) and let R to show that f (n) (z) = 0 for all z.

69

10. Show that if f : C C is holomorphic and not constant, then its range f (C) is dense
in C.
Hint. If f (C) omits a neighborhood of p C, consider the holomorphic function g(z) =
1/(f (z) p).
11. Consider the functions
f (z) = ez z,

f 0 (z) = ez 1.

Show that all the zeros of f are contained in {z : Re z > 0} while all the zeros of f 0 lie on
the imaginary axis. (Contrast this with the result of Exercise 7.)
Remark. Result of 29 imply that ez z has infinitely many zeros.

70

7. Harmonic functions on planar regions

We can write the Cauchy-Riemann equation (1.28) as (/x + i/y)f = 0. Applying
/x i/y to this gives
(7.1)

2f
2f
+
= 0,
x2
y 2

on an open set C, whenever f is holomorphic on . In general, a C 2 solution to

(7.1) on such is called a harmonic function. More generally, if O is an open set in Rn , a
function f C 2 (O) is said to be harmonic on O if f = 0 on O, where
(7.2)

f =

2f
2f
+

+
.
x21
x2n

Here we restrict attention to the planar case, n = 2. Material on harmonic functions

in higher dimensions can be found in many books on partial differential equations, for
example [T2] (particularly in Chapters 3 and 5), and also in Advanced Calculus texts,
such as [T] (see 10).
If f = u + iv is holomorphic, with u = Re f, v = Im f , (7.1) implies
(7.3)

2u 2u
+ 2 = 0,
x2
y

2v
2v
+
= 0,
x2
y 2

so the real and imaginary parts of a function holomorphic on a region are both harmonic
on . Our first task in this section will be to show that many (though not all) domains
C have the property that if u C 2 () is real valued and harmonic, then there exists
a real valued function v C 2 () such that f = u + iv is holomorphic on . One says
v is a harmonic conjugate to u. Such a property is equivalent to the form (1.34) of the
Cauchy-Riemann equations:
(7.4)

u
v
=
,
x
y

v
u
= .
x
y

To set up a construction of harmonic conjugates, we fix some notation. Given =

a + ib, z = x + iy (a, b, x, y R), let z denote the path from to z consisting of the
vertical line segment from a + ib to a + iy, followed by the horizontal line segment from
a + iy to x + iy. Let z denote the path from to z consisting of the horizontal line
segment from a + ib to x + ib, followed by the vertical line segment from x + ib to x + iy.
Also, let Rz denote the rectangle bounded by these four line segments. See Fig. 7.1. Here
is a first construction of harmonic conjugates.

71

Proposition 7.1. Let C be open, = a + ib , and assume the following property

holds:
(7.5)

If also z , then Rz .

Let u C 2 () be harmonic. Then u has a harmonic conjugate v C 2 ().

Proof. For z , set

(7.6)

Z
u
u
v(z) =
dx +
dy

y
x
z
Z y
Z x
u
u
=
(a + is) ds
(t + iy) dt.
b x
a y

Also set

(7.7)

Z
u
u
v(z) =

dx +
dy
y
x
z
Z x
Z y
u
u
=
(t + ib) dt +
(x + is) ds.
a y
b x

(7.8)

v
u
(z) = (z),
x
y

and
(7.9)

v
u
(z) =
(z).
y
x

Furthermore, since Rz , we have

Z
u
u
v(z) v(z) =

dx +
dy
y
x
Rz

(7.10)

ZZ 2
u 2u
=
+ 2 dx dy
x2
y
Rz

= 0,
the second identity by Greens theorem, (5.1), and the third because u = 0 on . Hence
(7.8)(7.9) give the Cauchy-Riemann equations (7.4), proving Proposition 7.1.
The next result, whose proof is similar to that of Proposition 1.10, simultaneously
extends the scope of Proposition 7.1 and avoids the use of Greens theorem.

72

Proposition 7.2. Let C be open, = a + ib , and assume the following property

holds.
(7.11)

If also z , then z .

Let u C 2 () be harmonic. Then u has a harmonic conjugate v C 2 ().

Proof. As in the proof of Proposition 7.1, define v on by (7.6). We again have (7.8). It
remains to compute v/y. Applying /y to (7.6) gives
Z x 2
v
u
u
(z) =
(a + iy)
(t + iy) dt
2
y
x
a y
Z x 2
u
u
(a + iy) +
=
(t + iy) dt
2
x
a t
(7.12)
x
u
u

(a + iy) +
(t + iy)
=
x
x
t=a
u
=
(z),
x
the second identity because u is harmonic and the third by the fundamental theorem of
calculus. This again establishes the Cauchy-Riemann equations, and completes the proof
of Proposition 7.2.
Later in this section, we will establish the existence of harmonic conjugates for a larger
class of domains. However, the results given above are good enough to yield some important
information on harmonic functions, which we now look into. The following is the mean
value property for harmonic functions.
Proposition 7.3. If u C 2 () is harmonic, z0 , and Dr (z0 ) , then
Z 2
1
(7.13)
u(z0 ) =
u(z0 + rei ) d.
2 0
Proof. We can assume u is real valued. Take > r such that D (z0 ) . By Proposition
7.1, u has a harmonic conjugate v on D (z0 ), so f = u + iv is holomorphic on D (z0 ). By
(5.8),
Z 2
1
(7.14)
f (z0 ) =
f (z0 + rei ) d.
2 0
Taking the real part gives (7.13).
As in (6.3), we also have, under the hypotheses of Proposition 7.3,
ZZ
Z 2 Z r
u(z0 + sei )s ds d
u(z) dx dy =
0
0
(7.15)
Dr (z0 )
= r2 u(z0 ),

73

hence
(7.16)

1
u(z0 ) =
Ar

ZZ
u(z) dx dy.

Dr (z0 )

With this, we can establish a maximum principle for harmonic functions.

Proposition 7.4. Let C be a connected, open set. If u : R is harmonic on ,
then, given z0 ,
(7.17)

z

(7.18)

z

Proof. Essentially the same as the proof of Proposition 6.1.

Next, we establish Liouvilles theorem for harmonic functions on C.
Proposition 7.5. If u C 2 (C) is bounded and harmonic on all of C, then u is constant.
Proof. Pick any two points p, q C. We have, for all r > 0,
ZZ
ZZ
i
1 h
u(z) dx dy
u(z) dx dy ,
(7.19)
u(p) u(q) =
Ar
Dr (p)

Dr (q)

(7.20)

1
|u(p) u(q)| 2
r

ZZ
|u(z)| dx dy,

(p,q,r)

where
(7.21)

(7.22)

(7.23)

|u(p) u(q)|

4M |p q|
,
r

r < ,

74

and taking r gives u(p) u(q) = 0, so u is constant.

Second proof of Proposition 7.5. Take u as in the statement of Proposition 7.5. By Proposition 7.1, u has a harmonic conjugate v, so f (z) = u(z) + iv(z) is holomorphic on C, and,
for some M < ,
| Re f (z)| M, z C.
In such a case, Re(f (z) + M + 1) 1 for all z, so
g(z) =

1
f (z) + M + 1

is holomorphic on C and |g(z)| 1 for all z, so Proposition 6.3 implies g is constant, which
implies f is constant. (Compare Exercise 1 in 6.)
We return to the question of when does a harmonic function on a domain C have a
harmonic conjugate. We start with a definition. Let C be a connected, open set. We
say is a simply connected domain if the following property holds. Given p, q and a
pair of smooth paths
(7.24)

0 , 1 : [0, 1] ,

j (0) = p, j (1) = q,

(7.25)

s : [0, 1] ,

s (0) = p, s (1) = q, s [0, 1].

Compare material in Exercises 1013 of 5. The definition given there looks a little different, but it is equivalent to the one given here. We also write s (t) = (s, t), :
[0, 1] [0, 1] . We will prove the following.
Proposition 7.6. Let C be a simply connected domain. Then each harmonic function
u C 2 () has a harmonic conjugate.
The proof starts like that of Proposition 7.1, picking and setting
Z
u
u
(7.26)
v(z) =

dx +
dy ,
y
x
z

except this time, z denotes an arbitrary piecewise smooth path from to z. The crux
of the proof is to show that (7.26) is independent of the choice of path from to z. If this
known, we can simply write
Z z
u
u
(7.27)
v(z) =

dx +
dy ,
y
x

and proceed as follows. Given z , take r > 0 such that Dr (z) . With z = x + iy,
pick + iy, x + i Dr (z) (x, y, , R). We have
Z +iy
Z x+iy
u
u
u
u
(7.28)
v(z) =

dx +
dy +

dx +
dy ,
y
x
y
x

+iy

75

where we go from + iy to z = x + iy on a horizontal line segment, and a calculation

parallel to (7.8) gives
v
u
(z) = (z).
x
y

(7.29)
We also have
Z
(7.30)

x+i

v(z) =

u
u

dx +
dy +
y
x

x+iy

x+i

u
u
dx +
dy ,
y
x

to (7.9) gives
(7.31)

v
u
(z) =
(z),
y
x

thus establishing that v is a harmonic conjugate of u.

It remains to prove the asserted path independence of (7.26). This is a special case of
the following result.
Lemma 7.7. Let C be a simply connected domain, and pick p, q . Assume
F1 , F2 C 1 () satisfy
F1
F2
=
.
y
x

(7.32)
Then
Z
(7.33)

F1 dx + F2 dy
pq

is independent of the choice of path from p to q.

To see how this applies to the path independence in (7.26), which has the form (7.33)
with
(7.34)

F1 =

u
,
y

F2 =

u
,
x

(7.35)

F1
2u
= 2,
y
y

F2
2u
=
,
x
x2

so (7.32) is equivalent to the assertion that u is harmonic on .

In view of the definition of a simply connected domain, the following result implies
Lemma 7.7.

76

Lemma 7.8. Let C be open and connected, and pick p, q . Let F1 , F2 C 1 ()

satisfy (7.32). Then, if s : [0, 1] , 0 s 1, is a smooth family of paths from p to q,
Z
(7.36)
F1 dx + F2 dy
s

is independent of s.
To prove Lemma 7.8, it is convenient to introduce some notation. Set
(7.37)

x1 = x,

x2 = y,

so (7.32) yields
Fk
Fj
=
,
xj
xk

(7.38)

for all j, k {1, 2}. Also, represent elements of R2 C as vectors:

0
X
F1
1
0
0
(7.39)
F =
, =
,
F

=
Fj j0 .
F2
20
j
Then (7.36) takes the form
Z
(7.40)
0

We are assuming

s (t) = (s, t),

: [0, 1] [0, 1] ,

(s, 0) = p,

(s, 1) = q,

and the claim is that (7.40) is independent of s, provided (7.38) holds.

To see this independence, we compute the s-derivative of (7.40), i.e., of
Z 1

F ((s, t))
(s, t) dt
t
0
Z 1X
(7.42)
j
Fj ((s, t))
=
(s, t) dt.
t
0
j
The s-derivative of the integrand in (7.42) is obtained via the product rule and the chain
rule. Thus the s-derivative of (7.42) is
Z 1X
Fj

s
t
0 j,k xk
(7.43)
Z 1X

+
j (s, t) dt.
Fj ((s, t))
s t
0
j

77

We can apply the identity

j (s, t) =
j (s, t)
s t
t s

(7.44)

to the second integrand in (7.43) and then integrate by parts. This involves applying /t
to Fj ((s, t)), and hence another application of the chain rule. When this is done, the
second integral in (7.43) becomes
Z 1X
Fj

(7.45)

xk
t
s
0
j,k

Note that (s, 0) p, (s, 1) q s j (s, 0) 0 s (s, 1), so integration by parts

involves no endpoint contributions. Now, if we interchange the roles of j and k in (7.45),
we cancel the first integral in (7.43), provided (7.38) holds. This proves Lemma 7.8.
With this done, the proof of Proposition 7.6 is complete.

Exercises
1. Modify the second proof of Proposition 7.5 to establish the following.
Proposition 7.5A. If u C 2 (C) is harmonic on C, real valued, and bounded from below,
then u is constant.
Hint. Let u = Re f, f : C C, holomorphic. See Exercise 1 of 6.
Alternative. Modify the argument involving (7.19)(7.23), used in the first proof of Proposition 7.5, to show that u(p) u(q) 0, and then reverse the roles of p and q.
2. Let u : R be harmonic. Assume u has a harmonic conjugate v, so f (z) = u(z)+iv(z)
is holomorphic. Show that, if : [a, b] is a piecewise smooth path,
Z
Z
Z
1
u
u
v
v
0
(7.46)
f (z) dz =

dx +
dy + i

dx +
dy .
i
y
x
y
x

Deduce that the integral on the right side of (7.46) must vanish whenever is a closed
curve in .
Hint. Write f 0 (z) dz = (ux + ivx )(dx + idy), separate the left side of (7.46) into real and
imaginary parts, and use the Cauchy-Riemann equations to obtain the right side of (7.46).
3. Let C be a connected, open set and u : R be harmonic. Show that u has a
harmonic conjugate v C 2 () if and only if
Z
u
u
(7.47)

dx +
dy = 0,
y
x

78

for every smooth closed curve in .

Hint. For the only if part, use (7.46). For the converse, consider the role of (7.26) in
the proof of Proposition 7.6.
Recall from 4 the holomorphic function log : C \ (, 0] C, characterized by
z = rei , r > 0, < < = log z = log r + i.
In particular, if we define Arg(z) to be , then log |z| is harmonic on C \ (, 0], with
harmonic conjugate Arg(z).
4. Show directly that log |z| is harmonic on C \ 0.
5. Show that log |z| does not have a harmonic conjugate on C \ 0.
One approach: Apply Exercise 3.
Hint. Show that if (t) = reit , 0 t 2, and u(z) = (|z|2 ), then
(7.48)

Z
u
u

dx +
dy = 4r2 0 (r2 ).
y
x

6. Let C be open and connected, u : R harmonic. If v and v are both harmonic

conjugates of u on , show that v v is constant.
Hint. Use the Cauchy-Riemann equations to show that if g : R is holomorphic, then
g is constant.
7. Use Exercise 6 to get another solution to Exercise 5, via the results on Arg(z).
8. Let C be simply connected. Pick p . Given g holomorphic on , set
Z
f (z) =

g() d,

z ,

pz

with pz a smooth path in from p to z. Show that this integral is independent of the
choice of such a path, so f (z) is well defined. Show that f is holomorphic on and
f 0 (z) = g(z),

z .

Hint. For the independence, see Exercise 14 of 5. For the rest, adapt the argument used
to prove Proposition 7.6, to show that
f
(z) = g(z),
x

and

1 f
(z) = g(z).
i y

79

9. Let C be a bounded domain with piecewise smooth boundary. Let F1 , F2 C 1 ()

satisfy (7.32). Use Greens theorem to show that
Z
F1 dx + F2 dy = 0.

Compare this conclusion with that of Lemma 7.7. Compare the relation between these
two results with the relation of Proposition 5.7 to Theorem 5.2.
10. Let C be open, and pq : [a, b] a path from p to q. Show that if v C 1 (),
then
Z
v
v
(7.49)
v(q) v(p) =
dx +
dy .
x
y
pq

Relate this to the use of (7.6), (7.7), and (7.26).

Hint. Compute (d/dt)v((t)).
11. Show that (7.49) implies Proposition 1.8.

80

8. Moreras theorem and the Schwarz reflection principle

Let be a connected open set in C. We have seen that if f : C is holomorphic,
i.e., f C 1 () and f is complex-differentiable, then the Cauchy integral theorem and the
Cauchy integral formula hold for f , and hence f C () and f 0 is also holomorphic. Here
we will establish a converse of the Cauchy integral theorem, known as Moreras theorem.
Theorem 8.1. Assume g : C is continuous and
Z
(8.1)
g(z) dz = 0

whenever = R and R is a rectangle. Then g is holomorphic.

Proof. Since the property of being holomorphic is local, there is no loss of generality in
assuming is a rectangle. Fix = a + ib . Given z = x + iy , let z and z
be the piecewise linear paths from to z described below (7.4); cf. Fig. 7.1. That is, z
goes vertically from a + ib to a + iy, then horizontally from a + iy to x + iy, and z goes
horizontally from a + ib to x + ib, then vertically from x + ib to x + iy. Now set
Z
Z y
Z x
(8.2)
f (z) =
g() d = i
g(a + is) ds +
g(t + iy) dt.
b

(8.2A)

f (z) =

Z
g() d =

g(s + ib) ds + i
a

b

Applying /x to (8.2) gives (as in (1.53))

f
(z) = g(z).
x
Similarly, applying /y to (8.2A) gives
(8.3)

(8.4)

f
(z) = ig(z).
y

This shows that f : C is C 1 and satisfies the Cauchy-Riemann equations. Hence f is

holomorphic and f 0 (z) = g(z). Thus g is holomorphic, as asserted.
Moreras theorem helps prove an important result known as the Schwarz reflection
principle, which we now discuss. Assume C is an open set that is symmetric about
the real axis, i.e.,
(8.5)

z = z .

81

Proposition 8.2. In the set-up above, assume f : + L C is continuous, holomorphic

in + , and real valued on L. Define g : C by
g(z) = f (z),

(8.6)

f (z),

z + L,
z .

Then g is holomorphic on .
Proof. It is readily verified that g is C 1 on and satisfies the Cauchy-Riemann equation
there, so g is holomorphic on \ L. Also it is clear that g is continuous on . To see that
g is holomorphic on all of , we show that g satisfies (8.1) whenever = R and R
is a (closed) rectangle. If R + or R this is clear. If R + L, it follows by
the continuity of g and a limiting argument (see Theorem 5.9); similarly we treat the case
R L. Finally, if R intersects both + and , then we set R = R+ R with
R = L, and note that
Z
Z
Z
g(z) dz,
g(z) dz =
g(z) dz +
R

R+

to finish the proof.

Remark. For a stronger version of the Schwarz reflection principle, see Proposition 13.9.

Exercises
1. Let C be a connected domain. Suppose is a smooth curve in , and \ has
two connected pieces, say . Assume g is continuous on , and holomorphic on + and
on . Show that g is holomorphic on .
Hint. Verify the hypotheses of Moreras theorem.
2. Suppose f is holomorphic on the semidisk |z| < 1, Im z > 0, continuous on its closure,
and real valued on the semicircle |z| = 1, Im z > 0. Show that setting
g(z) = f (z),

|z| 1, Im z > 0,

f (1/z),

defines a holomorphic function on the upper half-plane Im z > 0.

3. Suppose that f is holomorphic (and nowhere vanishing) on the annulus 1 < |z| < 2,
continuous on its closure, and |f | = 1 on the circle |z| = 1. Show that setting
g(z) = f (z),
1
f (1/z)

1 |z| 2,
,

82

defines a holomorphic function on the annulus 1/2 < |z| < 2.

4. The proof of Proposition 8.2 used the assertion that if f : + C is holomorphic and
on = {z C : z + }. Prove this.
g(z) = f (z), then g is holomorphic P
Hint. Given z0 + , write f (z) = k0 ak (z z0 )k on a neighborhood of z0 in + . Then
produce a power series for g about z 0 .
5. Given f : C, we say f is antiholomorphic if f is holomorphic, where f (z) = f (z).
Let g : O , with O and open in C. Prove the following:
(a) f holomorphic, g antiholomorphic = f g antiholomorphic.
(b) f antiholomorphic, g holomorphic = f g antiholomorphic.
(c) f antiholomorphic, g antiholomorphic = f g holomorphic.

83

9. Goursats theorem
If C is open and f : C, we have defined f to be holomorphic provided
f C 1 () and f is complex-differentiable. Here we show that the hypothesis f C 1 ()
can be dispensed with. This observation is due to E. Goursat.
Theorem 9.1. If f : C is complex-differentiable at each point of , then f is
holomorphic, so f C 1 (), and in fact f C ().
Proof. We will show that the hypothesis yields
Z
(9.1)
f (z) dz = 0
R

for every rectangle R . The conclusion

then follows from Moreras theorem.
R
Given a rectangle R , set a = R f (z) dz. Divide R into 4 equal rectangles. The
integral of f (z) dz over their boundaries sums to a. Hence one of them (call it R1 ) must
have the property that
(9.2)

Z
|a|

.
f
(z)
dz

4
R1

Divide R1 into four equal rectangles. One of them (call it R2 ) must have the property that
(9.3)

f (z) dz 42 |a|.

R2

such that
Z

(9.4)
f (z) dz 4k |a|.

Rk

have
(9.5)

with
(9.6)

84

In particular,
(9.7)

sup
zRk

|(z)|
= k 0, as k .
|z p|

Z
(9.8)

Z
dz = 0,

Rk

z dz = 0.
Rk

Hence, with k as in (9.7),

(9.9)

Z
Z

f (z) dz =
(z) dz Ck 2k 2k ,

Rk

Rk

since |z p| C 2k for z Rk , and the length of Rk is C 2k . Comparing (9.4) and

(9.9), we see that |a| Ck for all k, and hence a = 0. This establishes (9.1) and hence
proves Goursats theorem.

Exercise
1. Define f : R C by

1
f (x) = x2 sin , x 6= 0,
x
0,
x = 0.

Show that f is differentiable on R, but f 0 is not continuous on R. Contrast this with

Theorem 9.1.
2. Define f : C \ 0 C by

1
f (z) = z 2 sin ,
z

z 6= 0.

85

10. Uniqueness and analytic continuation

It is a central fact that a function holomorphic on a connected open set C is
uniquely determined by its values on any set S with an accumulation point in , i.e., a
point p with the property that for all > 0, the disk D (p) contains infinitely many
points in S. Phrased another way, the result is:
Proposition 10.1. Let C be open and connected, and let f : C be holomorphic.
If f = 0 on a set S and S has an accumulation point p , then f = 0 on .
Proof. There exists R > 0 such that the disk DR (p) and f has a convergent power
series on DR (p):
(10.1)

f (z) =

an (z p)n .

n=0

If all an = 0, then f = 0 on DR (p). Otherwise, say aj is the first nonzero coefficient, and
write
(10.2)

f (z) = (z p) g(z),

g(z) =

aj+n (z p)n .

n=0

Now g(p) = aj 6= 0, so there is a neighborhood U of p on which g is nonvanishing. Hence

f (z) 6= 0 for z U \ p, contradicting the hypothesis that p is an accumulation point of S.
This shows that if S # is the set of accumulation points in of the zeros of f , then S #
is open. It is elementary that S # is closed, so if is connected and S # 6= , then S # = ,
which implies f is identically zero.
To illustrate the use of Proposition 10.1, we consider the following Gaussian integral:
Z
(10.3)

G(z) =

et

+tz

dt.

It is easy to see that the integral is absolutely convergent for each z C and defines a
continuous function of z. Furthermore, if is any closed curve on C (such as = R for
some rectangle R C) then we can interchange order of integrals to get
Z
(10.4)

Z
et

G(z) dz =

+tz

dz dt = 0,

the last identity by Cauchys integral theorem. Then Moreras theorem implies that G is
holomorphic on C.

86

For z = x real we can calculate (10.3) via elementary calculus. Completing the square
in the exponent gives
Z
Z
2
x2 /4
(tx/2)2
x2 /4
(10.5)
G(x) = e
e
dt = e
et dt.

Z
2

(10.6)

R
0

I =

where

s2 t2

ds dt =

er r dr d = ,

(10.7)

G(x) =

/4

x R.

/4

z C,

ex

, so

(10.8)

G(z) =

ez

since both sides are holomorphic on C and coincide on R. In particuler, G(iy) =

for y R, so we have
Z

2
2
(10.9)
et +ity dt = ey /4 , y R.

ey

/4

We next prove the following

Proposition 10.2. Let C be a simply connected domain. Assume f : C is
holomorphic and nowhere vanishing. Then there exists a holomorphic function g on
such that
(10.10)

eg(z) = f (z),

z .

Proof. We may as well assume f is not constant. Take p such that f (p)
/ R =
(, 0]. Let O be a neighborhood of p such that
(10.11)

f : O C \ R .

We can define g on O by
(10.12)

g(z) = log f (z),

z O,

where log : C \ R C is as in 4. Applying d/dz and using the chain rule gives
(10.13)

g 0 (z) =

f 0 (z)
,
f (z)

z O.

87

Consequently
Z
(10.14)

p

f 0 ()
d,
f ()

for z O, where we integrate along a path from p to z within O. Now if f is nowhere

vanishing, then f 0 /f is holomorphic on , and if is simply connected, then the integral
on the right side is well defined for all z and is independent of the choice of path from
p to z, within , and defines a holomorphic function on . (See Exercise 14 of 5 and
Exercise 8 of 7.)
Hence (10.14) gives a well defined holomorphic function on . From (10.12), we have
(10.15)

eg(z) = f (z),

z O,

and then Proposition 10.1 implies (10.10).

It is standard to denote the function produced above by log f (z), so
Z z 0
f ()
(10.16)
log f (z) = log f (p) +
d,
p f ()
under the hypotheses of Proposition 10.2. One says that log f (z) is extended from O to
by analytic continuation. In the setting of Proposition 10.2, we can set
p
(10.17)
f (z) = e(1/2) log f (z) , z ,
and, more generally, for a C,
(10.18)

f (z)a = ea log f (z) ,

z ,

These functions are hence analytically continued from O (where they have a standard
definition from 4) to . We will see further cases of analytic continuation in Sections 18
and 19.

Exercises
1. Suppose C is a connected region that is symmetric about the real axis, i.e.,
z z . If f is holomorphic on and real valued on R, show that
(10.19)

f (z) = f (z).

Hint. Both sides are holomorphic. How are they related on R?

1A. Set z = z and note that z 7 z is reflection about the imaginary axis, just as z 7 z

88

is reflection about the real axis. Suppose C is a connected region that is symmetric
about the imaginary axis, i.e., z z . If f is holomorphic on and is purely
imaginary on iR, show that
f (z) = f (z ) .

(10.20)

Hint. Show that f (z ) = f (z) is holomorphic in z.

2. Let D be the unit disk centered at the origin. Assume f : D C is holomorphic
and that f is real on D R and purely imaginary on D iR. Show that f is odd, i.e.,
f (z) = f (z).
Hint. Show that (10.19) and (10.20) both hold.
3. Let be a simply connected domain in C and f a holomorphic function on with the
property that f : C \ {1, 1}. Assume f is not constant, and take p such that
f (p)
/ C \ {(, 1] [1, )}, the set where sin1 is defined; cf. (4.23)(4.27). Show that
sin1 f (z) can be analytically continued from a small neighborhood of p to all of , and
Z z
f 0 ()
1
1
p
sin f (z) = sin f (p) +
d.
1 f ()2
p
4. In the setting of Proposition 10.2, if g(z) is given by (10.14), show directly that

d g(z)
e
f (z) = 0,
dz
and deduce that (10.10) holds, without appealing to Proposition 10.1.
5. Consider

Z
I(a) =

eat dt,

Re a > 0.

Show that I is holomorphic in {a C : Re a > 0}. Show that

1/2
I(a) =
a
.
2
Hint. Use a change of variable to evaluate I(a) for a (0, ).
6. Evaluate

b > 0.

Make the evaluation explicit at b = 1.

Hint. Evaluate I(b i). Consult Exercise 19 of 4.
7. Show that

i/4
lim I(b i) =
e
.
b&0
2
R
R
What does this say about 0 cos t2 dt and 0 sin t2 dt?

89

11. Singularities
The function 1/z is holomorphic on C \ 0 but it has a singularity at z = 0. Here we will
make a further study of singularities of holomorphic functions.
A point p C is said to be an isolated singularity of f if there is a neighborhood U of p
such that f is holomorphic on U \ p. The singularity is said to be removable if there exists
a holomorphic function f on U such that f = f on U \ p. Clearly 0 is not a removable
singularity for 1/z, since this function is not bounded on any set D (0) \ 0. This turns out
to be the only obstruction to removability, as shown in the following result, known as the
removable singularities theorem.
Theorem 11.1. If p and f is holomorphic on \p and bounded, then p is a removable
singularity.
Proof. Consider the function g : C defined by
g(z) = (z p)2 f (z),

(11.1)

z \ p,

g(p) = 0.

(11.2)

z \ p,

g (p) = 0.

Thus (by Goursats theorem) g is holomorphic on , so on a neighborhood U of p it has

a convergent power series:
(11.3)

g(z) =

an (z p)n ,

z U.

n=0

(11.4)

h(z) =

a2+n (z p)n ,

z U.

n=0

(11.5)

f(z) = f (z),

z \ p,

f(p) = h(p)

90

By definition an isolated singularity p of a holomorphic function f is a pole if |f (z)|

as z p. Say f is holomorphic on \ p with pole at p. Then there is a disk U centered at
p such that |f (z)| 1 on U \ p. Hence g(z) = 1/f (z) is holomorphic on U \ p and g(z) 0
as z p. Thus p is a removable singularity for g. Let us also denote by g the holomorphic
extension, with g(p) = 0. Thus g has a convergent power series expansion valid on U :

X
(11.6)
g(z) =
an (z p)n ,
n=k

where we have picked ak as the first nonzero coefficient in the power series. Hence
(11.7)

h(p) = ak 6= 0,

with h holomorphic on U . This establishes the following.

Proposition 11.2. If f is holomorphic on \ p with a pole at p, then there exists k Z+
such that
(11.8)

on \ p, with F holomorphic on and F (p) 6= 0.

If Proposition 11.2 works with k = 1, we say f has a simple pole at p.
An isolated singularity of a function that is not removable and not a pole is called an
essential singularity. An example is
(11.9)

f (z) = e1/z ,

for which 0 is an essential singularity. The following result is known as the CasoratiWeierstrass theorem.
Proposition 11.3. Suppose f : \ p C has an essential singularity at p. Then, for
any neighborhood U of p in , the image of U \ p under f is dense in C.
Proof. Suppose that, for some neighborhood U of p, the image of U \ p under f omits a
neighborhood of w0 C. Replacing f (z) by f (z) w0 , we may as well suppose w0 = 0.
Then g(z) = 1/f (z) is holomorphic and bounded on U \ p, so p is a removable singularity
for g, which hence has a holomorphic extension g. If g(p) 6= 0, then p is removable for f .
If g(p) = 0, then p is a pole of f .
Remark. There is a strengthening of Proposition 11.3, due to E. Picard, which we will
treat in 28.
A function holomorphic on except for a set of poles is said to be meromorphic on .
For example,
sin z
tan z =
cos z
is meromorphic on C, with poles at {(k + 1/2) : k Z}.
Here we have discussed isolated singularities. In 4 we have seen examples of functions,
such as z 1/2 and log z, with singularities of a different nature, called branch singularities.
Another useful consequence of the removable singularities theorem is the following characterization of polynomials.

91

Proposition 11.4. If f : C C is holomorphic and |f (z)| as |z| , then f (z)

is a polynomial.
Proof. Consider g : C \ 0 C, defined by
(11.10)

g(z) = f

1
z

The hypothesis on f implies |g(z)| as z 0, so g has a pole at 0. By Proposition

11.2, we can write
g(z) = z k G(z)

(11.11)

on C \ 0, for some k Z+ , with G holomorphic on C and G(0) 6= 0. Then write

(11.12)

G(z) =

k1
X

gj z j + z k h(z),

j=0

with h(z) holomorphic on C. Then

g(z) =

k1
X

gj z jk + h(z),

j=0

so
f (z) =

k1
X

gj z kj + h

1
z

j=0

for z 6= 0.

It follows that
(11.13)

f (z)

k1
X

gj z kj

j=0

is holomorphic on C and, as |z| , this tends to the finite limit h(0). Hence, by
Liouvilles theorem, this difference is constant, so f (z) is a polynomial.

Exercises
1. In the setting of Theorem 11.1, suppose |z p| = r and Dr (z) . If |f | M on \ p,
show that
(11.14)

|f 0 (z)|

M
,
r

92

(11.15)

|(z p)2 f 0 (z)| M |z p|.

Use this to show directly that the hypotheses of Theorem 11.1 imply g C 1 (), avoiding
the need for Goursats theorem in the proof.
Hint. Use (5.10) to prove (11.14). In (5.10), replace by Ds (z) with s < r, and then let
s % r.
2. For yet another approach to the proof of Theorem 11.1, define h : C by
(11.16)

h(z) = (z p)f (z),

z \ p,

h(p) = 0.

R
Show that h : C is continuous. Show that R h(z) dz = 0 for each closed rectangle
R , and deduce that h is holomorphic on . Use a power series argument parallel to
that of (11.3)(11.4) to finish the proof of Theorem 11.1.
3. Suppose , O C are open and f : O is holomorphic and a homeomorphism.
Show that f 0 (p) 6= 0 for all p .
Hint. Apply the removable singularities theorem to f 1 : O .
Compare the different approach suggested in Exercise 8 of 5.
4. Let f : C C be holomorphic and assume f is not a polynomial. (We say f is a
transcendental function.) Show that g : C \ 0 C, defined by g(z) = f (1/z), has an
essential singularity at 0. Apply the Casorati-Weierstrass theorem to g, and interpret the
conclusion in terms of the behavior of f .
Remark. (Parallel to that after Proposition 11.3) For an improvement of this conclusion,
see 28.

93

12. Laurent series

There is a generalization of the power series expansion, which works for functions holomorphic in an annulus, rather than a disk. Let
(12.1)

A = {z C : r0 < |z z0 | < r1 }.

be such an annulus. For now assume 0 < r0 < r1 < . Let j be the counter-clockwise
circles {|z z0 | = rj }, so A = 1 0 . If f C 1 (A) is holomorphic in A, the Cauchy
integral formula gives
1
f (z) =
2i

(12.2)

Z
1

f ()
1
d
z
2i

f ()
d,
z

1
1
1
1
,
=
=
0
z
( z0 ) (z z0 )
z0 1 zz
z0

1 ,

1
1
,
z0
z z0 1 zz
0

0 ,

and use the fact that

|z z0 | < | z0 |,

for 1 ,

> | z0 |,

for 0 ,

to write
(12.3)

1
1 X z z0 n
=
,
z
z0 n=0 z0

1 ,

and
(12.4)

1
1 X z0 m
=
,
z
z z0 m=0 z z0

0 .

Plugging these expansions into (12.2) yields

(12.5)

f (z) =

X
n=

an (z z0 )n ,

z A,

94

with
(12.6)

1
an =
2i

Z
1

f ()
d,
( z0 )n+1

n 0,

and
(12.7)

1
an =
2i

Z
f ()( z0 )m d,

n = m 1 < 0.

Now in (12.6) and (12.7) 0 and 1 can be replaced by any circle in A concentric with
these. Using this observation, we can prove the following.
Proposition 12.1. Given 0 r0 < r1 , let A be the annulus (12.1). If f : A C is
holomorphic, then it is given by the absolutely convergent series (12.5), with
(12.8)

1
an =
2i

f ()
d,
( z0 )n+1

n Z,

where is any (counter-clockwise oriented) circle centered at z0 , of radius r (r0 , r1 ).

Proof. The preceding argument gives this result on every annulus
Ab = {z C : r00 < |z z0 | < r10 }
for r0 < r00 < r10 < r1 , which suffices.
Of particular interest is the case r0 = 0, dealing with an isolated singularity at z0 .
Proposition 12.2. Suppose f is holomorphic on DR (z0 ) \ z0 , with Laurent expansion
(12.5). Then f has a pole at z0 if and only if an = 0 for all but finitely many n < 0 (and
an 6= 0 for some n < 0). Hence f has an essential singularity at z0 if and only if an 6= 0
for infinitely many n < 0.
Proof. If z0 is a pole, the stated conclusion about the Laurent series expansion follows
from Proposition 11.2. The converse is elementary.
We work out Laurent series expansions for the function
(12.9)

f (z) =

1
,
z1

on the regions
(12.10)

{z : |z| < 1}

95

X
1
f (z) =
=
zk ,
1z

(12.11)

k=0

(12.12)

1 1
f (z) =
z 1

1
z

1
X
1 X 1 k
zk .
=
=
z
z
k=0

k=

Similarly, for
(12.13)

g(z) =

1
z2

on the regions
(12.14)

{z : |z| < 2}

(12.15)

1 X z k
=
,
2
2

1 1
g(z) =
2 1

z
2

k=0

and on the second region,

(12.16)

g(z) =

1 1
z 1

2
z

1
X
1 X 2 k
=
2k1 z k .
z
z
k=0

k=

Exercises
Consider the Laurent series
ez+1/z =

(12.17)

an z n .

n=

2. Show that

1
an =

Z
0

96

3. Show that, for k 0,

ak = ak =

1
.
m!(m
+
k)!
m=0

Hint. For Exercise 2, use (12.7); for Exercise 3 use the series for ez and e1/z .
4. Consider the function
f (z) =

1
.
(z 1)(z 2)

Give its Laurent series about z = 0:

a) on {z : |z| < 1},
b) on {z : 1 < |z| < 2},
c) on {z : 2 < |z| < }.
Hint. Use the calculations (12.9)(12.16).
5. Let C be open, z0 and let f be holomorphic on \ z0 and bounded. By
Proposition 12.1, we have a Laurent series
f (z) =

an (z z0 )n ,

n=

valid on {z : 0 < |z z0 | < b} for some b > 0. Use (12.8), letting shrink, to show that
an = 0 for each n < 0, thus obtaining another proof of the removable singularities theorem
(Theorem 11.1).
6. Show that, for |z| sufficiently small,

(12.18)

1 ez + 1
1 X
=
+
ak z 2k1 .
2 ez 1
z
k=1

F (z) =

1 cosh z/2
,
2 sinh z/2

F 0 (z) =

1
F (z)2 .
4

and show that

Using (2.19)(2.20), write out the Laurent expansion for F (z)2 , in terms of that for F (z)
given above. Comparing terms in the expansions of F 0 (z) and 1/4 F (z)2 , show that
a1 =

1
,
2

97

and, for k 2,
k1

1 X
ak =
a` ak` .
2k + 1
`=1

ak = (1)k1

Bk
,
(2k)!

and Bk are called the Bernoulli numbers.

7. As an alternative for Exercise 6, rewrite (12.18) as

X
X
z
z n X z `
2+
=
1+
ak z 2k ,
2
n!
`!
n=1
`=1

k=1

multiply out using (2.19)(2.20), and solve for the coefficients ak .

8. As another alternative, note that
1 ez + 1
1
1
= z
+ ,
z
2 e 1
e 1 2
and deduce that (12.18) is equivalent to
z=

X
z `
`=1

`!

z X
1 +
ak z 2k .
2

k=1

98

13. Fourier series and the Poisson integral

Given an integrable function f : S 1 C, we desire to write
(13.1)

f () =

f(k) eik ,

k=

for some coefficients f(k) C. We identify S 1 with R/(2Z). If (13.1) is absolutely

convergent, we can multiply both sides by ein and integrate. A change in order of
summation and integration is then justified, and using
Z
1
ei` d = 0 if ` 6= 0,
2
(13.2)

1 if ` = 0,
we see that
(13.3)

1
f(k) =
2

f ()eik d.

The series on the right side of (13.1) is called the Fourier series of f .
If f is given by (13.3), to examine whether (13.1) holds, we first sneak up on the sum
on the right side. For 0 < r < 1, set
(13.4)

Jr f () =

f(k)r|k| eik .

k=

Note that
(13.5)

1
|f(k)|
2

|f ()| d,

so whenever the right side of (13.5) is finite we see that the series (13.4) is absolutely
convergent for each r [0, 1). Furthermore, we can substitute (13.3) for f in (13.4) and
change order of summation and integration, to obtain
Z
(13.6)
Jr f () = f (0 )pr ( 0 ) d0 ,
S1

where

1 X |k| ik
pr () = p(r, ) =
r e
2

(13.7)
=

k=

h
X

1
1+
2

i
(rk eik + rk eik ) ,

k=1

99

and summing these geometrical series yields

1
1 r2
.
2 1 2r cos + r2
Let us examine p(r, ). It is clear that the numerator and denominator on the right side
of (13.8) are positive, so p(r, ) > 0 for each r [0, 1), S 1 . As r % 1, the numerator
tends to 0; as r % 1, the denominator tends to a nonzero limit, except at = 0. Since we
have
Z
Z X
1
(13.9)
p(r, ) d =
r|k| eik d = 1,
2
(13.8)

p(r, ) =

S1

we see that, for r close to 1, p(r, ) as a function of is highly peaked near = 0 and small
elsewhere, as in Fig. 13.1. Given these facts, the following is an exercise in real analysis.
Proposition 13.1. If f C(S 1 ), then
(13.10)

Jr f f uniformly on S 1 as r % 1.

Proof. We can rewrite (13.6) as

(13.11)

Jr f () =

f ( 0 )pr (0 ) d0 .

The results (13.8)(13.9) imply that for each (0, ),

Z
(13.12)
pr (0 ) d0 = 1 (r, ),
| 0 |

with (r, ) 0 as r % 1. Now, we break (13.11) into three pieces:

Z
Jr f () = f ()
pr (0 ) d0
Z

[f ( 0 ) f ()]pr (0 ) d0

Z
+
f ( 0 )pr (0 ) d0
+

(13.13)

| 0 |

= I + II + III.
We have
I = f ()(1 (r, )),
|II| sup |f ( 0 ) f ()|,

(13.14)

| 0 |

|III| (r, ) sup |f |.

These estimates yield (13.10).
From (13.10) the following is an elementary consequence.

100

Proposition 13.2. Assume f C(S 1 ). If the Fourier coefficients f(k) form a summable
series, i.e., if

(13.15)

|f(k)| < ,

k=

then the identity (13.1) holds for each S 1 .

P
Proof. What is to be shown is that if k |ak | < , then
(13.15A)

ak = S = lim

r%1

r|k| ak = S.

X

|ak | < .

|k|>N

Then

N
X

SN =

ak = |S SN | < ,

k=N

and

|k|
r ak < ,

r (0, 1).

|k|>N

Since clearly
lim

r%1

N
X
k=N

|k|

r ak =

N
X

ak ,

k=N

the conclusion in (13.15A) follows.

Remark. A stronger result, due to Abel, is that the implication (13.15A) holds without
the requirement of absolute convergence.
Note that if (13.15) holds, then the right side of (13.1) is absolutely and uniformly
convergent, and its sum is continuous on S 1 .
We seek conditions on f that imply (13.15). Integration by parts for f C 1 (S 1 ) gives,
for k 6= 0,
Z
1
i ik
f(k) =
f ()
(e
) d
2
k
(13.16)
Z
1
=
f 0 ()eik d,
2ik

101

hence
Z

1
|f(k)|
2|k|

(13.17)

|f 0 ()| d.

(13.18)

1
f(k) =
2k 2

hence

f 00 ()eik d,

1
|f(k)|
2k 2

|f 00 ()| d.

(13.19)

|f(k)|

2(k 2 + 1)
S1

Hence
(13.20)

f C 2 (S 1 ) =

|f(k)| < .

We will produce successive sharpenings of (13.20) below. We start with an interesting

example. Consider
(13.21)

f () = ||,

continued to be periodic of period 2. This defines a Lipschitz function on S 1 , whose

Fourier coefficients are
Z
1

f (k) =
|| eik d
2
(13.22)

1
= 1 (1)k
,
k 2
for k 6= 0, while f(0) = /2. It is clear this forms a summable series, so Proposition 13.2
implies that, for ,
|| =
(13.23)
=

X 2

eik
2
k 2
k odd

`=0

1
cos(2` + 1).
(2` + 1)2

102

We note that evaluating this at = 0 yields the identity

(13.24)

`=0

1
2
=
.
(2` + 1)2
8

Writing

X
1
k2

(13.25)

k=1

as a sum of two series, one for k 1 odd and one for k 2 even, yields an evaluation of
(13.25). (See Exercise 1 below.)
We see from (13.23) that if f is given by (13.21), then f(k) satisfies
|f(k)|

(13.26)

k2

C
.
+1

This is a special case of the following generalization of (13.20).

Proposition 13.3. Let f be Lipschitz continuous and piecewise C 2 on S 1 . Then (13.26)
holds.
Proof. Here we are assuming f is C 2 on S 1 \{p1 , . . . , p` }, and f 0 and f 00 have limits at each
of the endpoints of the associated intervals in S 1 , but f is not assumed to be differentiable
at the endpoints p` . We can write f as a sum of functions f , each of which is Lipschitz
on S 1 , C 2 on S 1 \ p , and f0 and f00 have limits as one approaches p from either side. It
suffices to show that each f (k) satisfies (13.26). Now g() = f ( + p ) is singular
only at = , and g(k) = f (k)eik(p ) , so it suffices to prove Proposition 13.3 when f
has a singularity only at = . In other words, f C 2 ([, ]), and f () = f ().
In this case, we still have (13.16), since the endpoint contributions from integration by
parts still cancel. A second integration by parts gives, in place of (13.18),
Z
1
i ik

f (k) =
f 0 ()
(e
) d
2ik
k
(13.27)
Z
i
1 h 00
ik
0
0
=
f
()e
d
+
f
()

f
()
,
2k 2
which yields (13.26).
Given f C(S 1 ), let us say
(13.28)

f A(S 1 )

|f(k)| < .

Proposition 13.2 applies to elements of A(S 1 ), and Proposition 13.3 gives a sufficient
condition for a function to belong to A(S 1 ). A more general sufficient condition will be
given in Proposition 13.6.
R
We next make use of (13.2) to produce results on S 1 |f ()|2 d, starting with the following.

103

(13.29)

1
|f(k)|2 =
2

Z
|f ()|2 d.
S1

More generally, if also g A(S 1 ),

(13.30)

1
f(k)
g (k) =
2

Z
f ()g() d.
S1

Proof. Switching order of summation and integration and using (13.2), we have
Z
Z X
1
1
f ()g() d =
f(j)
g (k)ei(jk) d
2
2
S1
S 1 j,k
(13.31)
X
=
f(k)
g (k),
k

giving (13.30). Taking g = f gives (13.29).

We will extend the scope of Proposition 13.4 below. Closely tied to this is the issue of
convergence of SN f to f as N , where
X
f(k)eik .
(13.32)
SN f () =
|k|N

Clearly f A(S 1 ) SN f f uniformly on S 1 as N . Here, we are interested in

convergence in L2 -norm, where
Z
1
2
(13.33)
kf kL2 =
|f ()|2 d.
2
S1

Given f and |f |2 integrable on S 1 (we say f is square integrable), this defines a norm,
satisfying the following result, called the triangle inequality:
(13.34)

kf + gkL2 kf kL2 + kgkL2 .

See Appendix H for details on this. Behind these results is the fact that
(13.35)

Z
1
f ()g() d.
(13.36)
(f, g)L2 =
2
S1

104

Thus the content of (13.29) is that

X

(13.37)

|f(k)|2 = kf k2L2 ,

and that of (13.30) is that

(13.38)

f(k)
g (k) = (f, g)L2 .

The left side of (13.37) is the square norm of the sequence (f(k)) in `2 . Generally, a
sequence (ak ) (k Z) belongs to `2 if and only if
(13.39)

k(ak )k2`2 =

|ak |2 < .

(13.40)

((ak ), (bk )) =

ak bk .

(13.41)

As for the notion of L2 -norm convergence, we say

(13.42)

f f in L2 kf f kL2 0.

(13.43)

In view of the uniform convergence SN f f for f A(S 1 ) noted above, we have

(13.44)

f A(S 1 ) = SN f f in L2 , as N .

The triangle inequality implies

(13.45)
kf kL2 kSN f kL2 kf SN f kL2 ,
and clearly (by Proposition 13.4)
(13.46)

kSN f k2L2

N
X
k=N

|f(k)|2 ,

105

so
kf SN f kL2 0 as N = kf k2L2 =

(13.47)

|f(k)|2 .

We now consider more general square integrable functions f on S 1 . With f(k) and SN f
defined by (13.3) and (13.32), we define RN f by
(13.48)
Note that

f = SN f + RN f.
R
S1

f ()eik d =

(13.49)

S1

(f, SN f )L2 = (SN f, SN f )L2 ,

and hence
(13.50)

(SN f, RN f )L2 = 0.

Consequently,
(13.51)

kf k2L2 = (SN f + RN f, SN f + RN f )L2

= kSN f k2L2 + kRN f k2L2 .

In particular,
(13.52)

We are now in a position to prove the following.

Lemma 13.5. Let f, f be square integrable on S 1 . Assume
(13.53)

lim kf f kL2 = 0,

and, for each ,

(13.54)

lim kf SN f kL2 = 0.

Then
(13.55)

lim kf SN f kL2 = 0.

In such a case, (13.47) holds.

Proof. Writing f SN f = (f f ) + (f SN f ) + SN (f f ), and using the triangle
inequality, we have, for each ,
(13.56)

106

(13.57)

N

for each . Then (13.53) yields the desired conclusion (13.55).

Given f C(S 1 ), we have seen that Jr f f uniformly (hence in L2 -norm) as r % 1.
Also, Jr f A(S 1 ) for each r (0, 1). Thus (13.44) and Lemma 13.5 yield the following.
f C(S 1 ) = SN f f in L2 , and
X
|f(k)|2 = kf k2L2 .

(13.58)

Lemma 13.5 also applies to many discontinuous functions. Consider, for example
f () = 0 for < < 0,
1 for 0 < < .

(13.59)
We can set, for N,

f () = 0

for 0,
1
for 0 ,

1
1
1 for

1
( ) for .

(13.60)

Then each f C(S 1 ). (In fact, f A(S 1 ), by Proposition 13.3.). Also, one can check
that kf f k2L2 2/. Thus the conclusion in (13.58) holds for f given by (13.59).
More generally, any piecewise continuous function on S 1 is an L2 limit of continuous
functions, so the conclusion of (13.58) holds for them. To go further, let us recall the class
of Riemann integrable functions. (Details can be found in Chapter 4, 2 of [T0] or in 0 of
[T].) A function f : S 1 R is Riemann integrable provided f is bounded (say |f | M )
and, for each > 0, there exist piecewise constant functions g and h on S 1 such that
Z
(13.61)

g f h ,

and

h () g () d < .

S1

Then
Z
(13.62)

Z
f () d = lim

S1

0
S1

Z
g () d = lim

0
S1

h () d.

107

Z
Z
1
M +1
2
|f () g ()| d
|h () g ()| d
2

S1
S1
(13.63)
M +1
<
,

so g f in L2 -norm. A function f : S 1 C is Riemann integrable provided its real and

imaginary parts are. In such a case, there are also piecewise constant functions f f in
L2 -norm, so
(13.64)

f Riemann interable on S 1 = SN f f in L2 , and

X
|f(k)|2 = kf k2L2 .

This is not the end of the story. There are unbounded functions on S 1 that are square
integrable, such as
1
(13.65)
f () = || on [, ], 0 < < .
2
In such a case, one can take f () = min(f (), ), N. Then each f is continuous and
kf f kL2 0 as . Hence the conclusion of (13.64) holds for such f .
The ultimate theory of functions for which the result
(13.66)

SN f f in L2 -norm

holds was produced by H. Lebesgue in what is now known as the theory of Lebesgue
measure and integration. There is the notion of measurability
of a function f : S 1
R
2
1
C. One says f L (S ) provided f is measurable and S 1 |f ()|2 d < , the integral
here being the Lebesgue integral. Actually,
L2 (S 1 ) consists of equivalence classes of such
R
functions, where f1 f2 if and only if |f1 () f2 ()|2 d = 0. With `2 as in (13.39), it
is then the case that
(13.67)

F : L2 (S 1 ) `2 ,

given by
(13.68)

is one-to-one and onto, with

X
(13.69)
|f(k)|2 = kf k2L2 ,

f L2 (S 1 ),

and
(13.70)

SN f f in L2 ,

f L2 (S 1 ).

For the reader who has not seen Lebesgue integration, we refer to books on the subject
We mention two key propositions which, together with the arguments given above,
establish these results. The fact that Ff `2 for all f L2 (S 1 ) and (13.69)(13.70) hold
follows via Lemma 13.5 from the following.

108

Proposition A. Given f L2 (S 1 ), there exist f C(S 1 ) such that f f in L2 .

As for the surjectivity of F in (13.67), note that, given (ak ) `2 , the sequence
f () =

ak eik

|k|

satisfies, for > ,

X

kf f k2L2 =

|ak |2 0 as .

<|k|

That is to say, (f ) is a Cauchy sequence in L2 (S 1 ). Surjectivity follows from the fact that
Cauchy sequences in L2 (S 1 ) always converge to a limit:
Proposition B. If (f ) is a Cauchy sequence in L2 (S 1 ), there exists f L2 (S 1 ) such
that f f in L2 -norm.
Proofs of these results can be found in the standard texts on measure theory and integration, such as [T3].
We now establish a sufficient condition for a function f to belong to A(S 1 ), more general
than that in Proposition 13.3.
P
Proposition 13.6. If f is a continuous, piecewise C 1 function on S 1 , then
|f (k)| < .
Proof. As in the proof of Proposition 13.3, we can reduce the problem to the case f
C 1 ([, ]), f () = f (). In such a case, with g = f 0 C([, ]), the integration by
parts argument (13.16) gives
1
f(k) = g(k),
ik

(13.71)

k 6= 0.

By (13.64),
X

(13.72)

|
g (k)|2 = kgk2L2 .

Also, by Cauchys inequality (cf. Appendix H),

X
(13.73)

k6=0

|f(k)|

X 1 1/2 X
1/2
2
|
g (k)|
k2
k6=0

k6=0

CkgkL2 .
This completes the proof.
There is a great deal more that can be said about convergence of Fourier series. For example, material presented in the appendix to the next section has an analogue for Fourier

109

series. We also mention Chapter 5, 4 in [T0]. For further results, one can consult treatments of Fourier analysis such as Chapter 3 of [T2].
Fourier series connects with the theory of harmonic functions, as follows. Taking z =
rei in the unit disk, we can write (13.4) as
(13.74)

Jr f () =

f(k)z k +

k=0

f(k)z k .

k=1

We write this as
(13.75)

a sum of a holomorphic function and a conjugate-holomorphic function on the unit disk

D. Thus the left side is a harmonic function, called the Poisson integral of f .
Given f C(S 1 ), PI f is the unique function in C 2 (D) C(D) equal to f on D = S 1
(uniqueness being a consequence of Proposition 7.4). Using (13.6)(13.8), we can write
the following Poisson integral formula:
(13.76)

1 |z|2
PI f (z) =
2

f (w)
ds(w),
|w z|2

S1
0

the integral being with respect to arclength on S 1 . To see this, note that if w = ei and
z = rei , then ds(w) = d0 and
0

|w z|2 = (ei rei )(ei rei )

0

= 1 r(ei( ) + ei( ) ) + r2 .
Since solutions to u = 0 remain solutions upon translation and dilation of coordinates,
we have the following result.
Proposition 13.7. If D C is an open disk and f C(D) is given, there exists a
unique u C(D) C 2 (D) satisfying
(13.77)

u = 0 on D,

uD = f.

We call (13.77) the Dirichlet boundary problem.

Now we make use of Proposition 13.7 to improve the version of the Schwarz reflection
principle given in Proposition 8.2. As in the discussion of the Schwarz reflection principle
in 8, we assume C is a connected, open set that is symmetric with respect to the real
axis, so z z . We set = {z : Im z > 0} and L = R.

110

Proposition 13.8. Assume u : + L C is continuous, harmonic on + , and u = 0

on L. Define v : C by
(13.78)

v(z) = u(z),
u(z),

z + L,
z .

Then v is harmonic on .
Proof. It is readily verified that v is harmonic in + and continuous on . We
need to show that v is harmonic on a neighborhood of each point p L. Let D = Dr (p)
be a disk centered at p such that D . Let f C(D) be given by f = v|D . Let
w C 2 (D) C(D) be the unique harmonic function on D equal to f on D.
Since f is odd with respect to reflection about the real axis, so is w, so w = 0 on D R.
+
Thus both v and w are harmonic on D+ = D {Im z > 0}, and continuous on D , and
+

agree on D+ , so the maximum principle implies w = v on D . Similarly w = v on D ,

and this gives the desired harmonicity of v.
Using Proposition 13.8, we establish the following stronger version of Proposition 8.2,
the Schwarz reflection principle, weakening the hypothesis that f is continuous on + L
to the hypothesis that Im f is continuous on + L (and vanishes on L). While this
improvement may seem a small thing, it can be quite useful, as we will see in 24.
Proposition 13.9. Let be as in Proposition 13.8, and assume f : + C is holomorphic. Assume Im f extends continuously to + L and vanishes on L. Define g : +
by
(13.79)

g(z) = f (z),

z + ,

f (z),

z .

Then g extends to a holomorphic function on .

Proof. It suffices to prove this under the additional assumption that is a disk. We
apply Proposition 13.8 to u(z) = Im f (z) on + , 0 on L, obtaining a harmonic extension
v : R. By Proposition 7.1, v has a harmonic conjugate w : R, so v + iw is
holomorphic, and hence h : C, given by
(13.80)

h(z) = w(z) + iv(z),

is holomorphic. Now Im h = Im f on + , so g h is real valued on + , so, being holomorphic, it must be constant. Thus, altering w by a real constant, we have
(13.81)

h(z) = g(z),

z + .

(13.82)

h(z) = h(z),

z .

111

(13.83)

z + ,

h(z) = g(z),

so h is the desired holomorphic extension.

Exercises
1. Verify the evaluation of the integral in (13.22). Use the evaluation of (13.23) at = 0
(as done in (13.24)) to show that
(13.84)

X
2
1
=
.
k2
6

k=1

f () = 1
0

for < < 0.

Then use (13.64) to obtain another proof of (13.84).

3. Apply (13.29) when f () is given by (13.21). Use this to show that

X
4
1
=
.
k4
90

k=1

4. Give the details for (13.76), as a consequence of (13.8).

5. Suppose f is holomorphic on an annulus containing the unit circle S 1 , with Laurent
series

X
f (z) =
an z n .
n=

Show that
an = g(n),

g = f S 1 .

Compare this with (12.8), with z0 = 0 and the unit circle S 1 .

Exercises 68 deal with the convolution of functions on S 1 , defined by
Z
1
f g() =
f ()g( ) d.
2
S1

112

6. Show that

h = f g = h(k)
= f(k)
g (k).

7. Show that
2

f, g L (S ), h = f g =

|h(k)|
< .

k=

Compute (k)

and (). Relate these computations to (13.21)(13.23).

9. Show that a formula equivalent to (13.76) is
1
PI f (z) =
2

(13.80)

ei + z
Re i
f () d.
e z

(We abuse notation by confounding f () and f (ei ), identifying S 1 and R/(2Z).)

10. Give the details for the improvement of Proposition 8.2 mentioned right after the proof
of Proposition 13.8. Proposition 7.6 may come in handy.
11. Let be symmetric about R, as in Proposition 13.8. Suppose f is holomorphic and
nowhere vanishing on + and |f (z)| 1 as z L. Show that f extends to be holomorphic
on , with |f (z)| = 1 for z L.
Hint. Consider the harmonic function u(z) = log |f (z)| = Re log f (z).
12. Given f () =

P
k

ak eik , show that f is real valued on S 1 if and only if

ak = ak ,

k Z.

13. Let f C(S 1 ) be real valued. Show that PI f and (1/i) PI g are harmonic conjugates,
provided
g(k) = f(k)
for k > 0,
f(k) for k < 0.

113

14. Fourier transforms

Take a function f that is integrable on R, so
Z
(14.1)
kf kL1 =
|f (x)| dx < .

We define the Fourier transform of f to be

Z
1

(14.2)
f () = Ff () =
f (x)eix dx,
2

R.

Similarly, we set
1
F f () =
2

(14.2A)

f (x)eix dx,

R,

and ultimately plan to identify F as the inverse Fourier transform.

Clearly
Z
1

(14.3)
|f ()|
|f (x)| dx.
2
We also have continuity.
Proposition 14.1. If f is integrable on R, then f is continuous on R.
R
Proof. Given > 0, pick N < such that |x|>N |f (x)| dx < . Write f = fN + gN ,
where fN (x) = f (x) for |x| N, 0 for |x| > N . Then
f() = fN () + gN (),

(14.4)
and

|
gN ()| < ,
2

(14.5)

Meanwhile, for , R,
(14.6)

1
fN () fN () =
2

and

(14.6A)

|eix eix | | | max eix

|x| | |
N | |,

114

for |x| N , so
N
|fN () fN ()| kf kL1 | |,
2

(14.7)

where kf kL1 is defined by (14.1). Hence each fN is continuous, and, by (14.5), f is a

uniform limit of continuous functions, so it is continuous.
The Fourier inversion formula asserts that
Z
1
(14.8)
f (x) =
f()eix d,
2
in appropriate senses, depending on the nature of f . We approach this in a spirit similar
to the Fourier inversion formula (13.1) of the previous section. First we sneak up on (14.8)
2
by inserting a factor of e . Set
Z
2
1
(14.9)
J f (x) =
f()e eix d.
2
2
Note that, by (14.3), whenever f L1 (R), f()e is integrable for each > 0. Furthermore, we can plug in (14.2) for f() and switch order of integration, getting
ZZ
2
1
J f (x) =
f (y)ei(xy) e dy d
2
Z
(14.10)
=
f (y)H (x y) dy,

where
1
H (x) =
2

(14.11)

+ix

d.

A change of variable shows that H (x) = (1/ )H1 (x/ ), and the computation of
H1 (x) is accomplished in 10; we see that H1 (x) = (1/2)G(ix), with G(z) defined by
(10.3) and computed in (10.8). We obtain
H (x) =

(14.12)
The computation

ex dx =

2
1
ex /4 .
4

done in (10.6) implies

(14.13)

H (x) dx = 1,

> 0.

We see that H (x) is highly peaked near x = 0 as & 0. An argument similar to that
used to prove Proposition 13.1 then establishes the following.

115

(14.14)

From here, parallel to Proposition 13.2, we have:

Corollary 14.3. Assume f is bounded and continuous on R, and f and f are integrable
on R. Then (14.8) holds for all x R.
Proof. If f L1 (R), then f is bounded and continuous. If also f L1 (R), then F f is
continuous. Furthermore, arguments similar to those used to prove Proposition 13.2 show
that the right side of (14.9) converges to the right side of (14.8) as & 0. That is to say,
(14.14A)

J f (x) F f(x),

as 0.

It follows from (14.14) that f (x) = F f(x).

Remark. With some more work, one can omit the hypothesis in Corollary 14.3 that f
be bounded and continuous, and use (14.14A) to deduce these properties as a conclusion.
This sort of reasoning is best carried out in a course on measure theory and integration.
At this point, we take the space to discuss integrable functions and square integrable
functions on R. Examples of integrable functions on R are bounded, piecewise continuous
functions satisfying (14.1). More generally, f could be Riemann integrable on each interval
[N, N ], and satisfy
Z N
(14.15)
lim
|f (x)| dx = kf kL1 < ,
N

where Riemann integrability on [N, N ] has a definition similar to that given in (13.61)
(13.62) for functions on S 1 . Still more general is Lebesgues class, consisting of measurable
functions f : R C satisfying (14.1), where the Lebesgue integral is used. An element of
1
we say f1 f2 provided
RL(R) consists of an equivalence class of such functions, where
1
|f f2 | dx = 0. The quantity kf kL1 is called the L norm of f . It satisfies the
1
triangle inequality
(14.16)

as an easy consequence of the pointwise inequality |f (x)+g(x)| |f (x)|+|g(x)| (cf. (0.14)).

Thus L1 (R) has the structure of a metric space, with d(f, g) = kf gkL1 . We say f f
in L1 if kf f kL1 0. Parallel to Propositions A and B of 13, we have the following.
Proposition A1. Given f L1 (R) and k N, there exist f C0k (R) such that f f
in L1 .
Here, C0k (R) denotes the space of functions with compact support whose derivatives of
order k exist and are continuous. There is also the following completeness result.

116

Proposition B1. If (f ) is a Cauchy sequence in L1 (R), there exists f L1 (R) such that
f f in L1 .
As in 13, we will also be interested in square integrable functions. A function f : R C
is said to be square integrable if f and |f |2 are integrable on each finite interval [N, N ]
and
Z
2
(14.17)
kf kL2 =
|f (x)|2 dx < .

Taking the square root gives kf kL2 , called the L2 -norm of f . Parallel to (13.34) and
(14.16), there is the triangle inequality
(14.18)

kf + gkL2 kf kL2 + kgkL2 .

The proof of (14.18) is not as easy as that of (14.16), but, like (13.34), it follows, via results
of Appendix H, from the fact that
(14.19)

where, for square integrable functions f and g,

Z
(14.20)

(f, g)L2 =

f (x)g(x) dx.

The triangle inequality (14.18) makes L2 (R) a metric space, with distance function d(f, g) =
kf gkL2 , and we say f f in L2 if kf f kL2 0. Parallel to Propositions A1 and
B1, we have the following.
Proposition A2. Given f L2 (R) and k N, there exist f C0k (R) such that f f
in L2 .
Proposition B2. If (f ) is a Cauchy sequence in L2 (R), there exists f L2 (R) such that
f f in L2 .
As in 13, we refer to books on measure theory, such as [T3], for further material on
L (R) and L2 (R), including proofs of results stated above.
Somewhat parallel to (13.28), we set
1

(14.21)

A(R) = {f L1 (R) : f bounded and continuous, f L1 (R)}.

By Corollary 14.3, the Fourier inversion formula (14.8) holds for all f A(R). It also
follows that f A(R) f A(R). Note also that
(14.22)

A(R) L2 (R).

117

In fact, if f A(R),
Z
kf k2L2
(14.23)

|f (x)|2 dx
Z
sup |f (x)| |f (x)| dx
=

1
kfkL1 kf kL1 .
2
It is of interest to know when f A(R). We mention one simple result here. Namely,
if f C k (R) has compact support (we say f C0k (R)), then integration by parts yields
(14.24)

0 j k.

Hence
(14.25)

C02 (R) A(R).

While (14.25) is crude, it will give us a start on the L2 -theory of the Fourier transform.
Let us denote by F the map f 7 f, and by F the map you get upon replacing eix by
eix . Then, with respect to the inner product (14.20), we have, for f, g A(R),
(14.26)

(14.27)

(14.28)

kFf kL2 = kF f kL2 = kf kL2 ,

for f, g A(R).
The result (14.28) is called the Plancherel identity. Using it, we can extend F and F
to act on L2 (R), obtaining (14.28) and the Fourier inversion formula on L2 (R).
Proposition 14.4. The maps F and F have unique continuous linear extensions from
(14.28A)

F, F : A(R) A(R)

to
(14.29)

F, F : L2 (R) L2 (R),

118

(14.30)

F Ff = f,

FF f = f

hold for all f L2 (R), as does (14.28).

This result can be proven using Propositions A2 and B2, and the inclusion (14.25),
which together with Proposition A2 implies that
(14.31)

For each f L2 (R), f A(R) such that f f in L2 .

The argument goes like this. Given f L2 (R), take f A(R) such that f f in L2 .
Then kf f kL2 0 as , . Now (14.28), applied to f f A(R), gives
(14.32)

as , . Hence (Ff ) is a Cauchy sequence in L2 (R). By Proposition B2, there exists

a limit h L2 (R); Ff h in L2 . One gets the same element h regardless of the choice of
(f ) such that (14.31) holds, and so we set Ff = h. The same argument applies to F f ,
which hence converges to F f . We have
(14.33)

kFf Ff kL2 , kF f F f kL2 0.

From here, the result (14.30) and the extension of (14.28) to L2 (R) follow.
Given f L2 (R), we have
[R,R] f f in L2 ,

as R ,

so Proposition 14.4 yields the following.

Proposition 14.5. Define SR by
(14.34)

1
SR f (x) =
2

f()eix d.

Then
(14.35)

f L2 (R) = SR f f in L2 (R),

as R .
Having Proposition 14.4, we can sharpen (14.25) as follows.

119

C01 (R) A(R).

(14.36)

Proof. Given f C01 (R), we can use (14.24) with j = k = 1 to get

g = f 0 = g() = i f().

(14.37)
Proposition 14.4 implies

kf + gkL2 = kf + f 0 kL2 .

(14.38)

Now, parallel to the proof of Proposition 13.6, we have

Z

(1 + i)1 (1 + i)f() d

kfkL1 =

nZ

Z
o1/2

d o1/2 n
()2 d

(1
+
i)
f
1 + 2

= kf + f 0 kL2 ,

(14.39)

the inequality in (14.39) by Cauchys inequality (cf. (H.18)) and the last identity by (14.37)
(14.38). This proves (14.36).
Remark. Parallel to Proposition 13.6, one can extend Proposition 14.6 to show that if
f has compact support, is continuuous, and is piecewise C 1 on R, then f A(R). In
conjunction with (14.39), the following is useful for identifying other elements of A(R).
Proposition 14.7. Let f A(R) and f C(R) L1 (R). Assume
f f in L1 -norm, and kf kL1 A,
for some A < . Then f A(R).
Proof. Clearly f f uniformly on R. Hence, for each R < ,
Z

|f () f()| d 0,

as .

Thus

|f()| d A,

R < ,

and it follows that f L1 (R), completing the proof.

120

The interested reader can consult 14A for a still sharper condition guaranteeing that
f A(R).

Exercises
In Exercises 12, assume f : R C is a C 2 function satisfying
|f (j) (x)| C(1 + |x|)2 ,

(14.40)

j 2.

1. Show that
|f()|

(14.41)

C0
,
2 + 1

R.

Deduce that f A(R).

2. With f given as in (14.1), show that

X
X
1
ikx

f (k)e
=
f (x + 2`).
2 k=
`=

(14.42)

This is known as the Poisson summation formula.

Hint. Let g denote the right side of (14.42), pictured as an element of C 2 (S 1 ). Relate the
Fourier series of g (`
a la 13) to the left side of (14.42).
2

3. Use f (x) = ex
(14.43)

/4t

r

X
1 X k2 /
`2
e
=
e
.

`=

k=

This is a Jacobi identity.

2
Hint. Use (14.11)(14.12) to get f() = 2t et . Take t = , and set x = 0 in (14.42).
4. For each of the following functions f (x), compute f().

(c)

f (x) = e|x| ,
1
f (x) =
,
1 + x2
f (x) = [1/2,1/2] (x),

(d)

(a)
(b)

Here I (x) is the characteristic function of a set I R. Reconsider the computation of

(b) when you get to 16.

121

5. In each case, (a)(d), of Exercise 4, record the identity that follows from the Plancherel
identity (14.28).
Exercises 68 deal with the convolution of functions on R, defined by
Z
(14.44)

f g(x) =

f (y)g(x y) dy.

6. Show that
kf gkL1 kf kL1 kgkL1 ,
7. Show that

\
(f
g)() =

sup |f g| kf kL2 kgkL2 .

2 f()
g ().

8. Compute f f when f (x) = [1/2,1/2] (x), the characteristic function of the interval
[1/2, 1/2]. Compare the result of Exercise 7 with the computation of (d) in Exercise 4.
9. Prove the following result, known as the Riemann-Lebesgue lemma.
(14.45)

f L1 (R) = lim |f()| = 0.

||

Hint. (14.41) gives the desired conclusion for f when f C02 (R). Then use Proposition
A1 and apply (14.3) to f f .
10. Sharpen the result of Exercise 1 as follows, using the reasoning in the proof of Proposition 14.6. Assume f : R C is a C 1 function satisfying
|f (j) (x)| C(1 + |x|)2 ,

j 1.

14A. More general sufficient condition for f A(R)

Here we establish a result substantially sharper than Proposition 14.6. We mention that
an analogous result holds for Fourier series. The interested reader can investigate this.
To set things up, given f L2 (R), let
(14.46)

fh (x) = f (x + h).

Our goal here is to prove the following.

122

Proposition 14.8. If f L1 (R) L2 (R) and there exists C < such that
kf fh kL2 Ch ,

(14.47)

h [0, 1],

with
(14.48)

>

1
,
2

then f A(R).
Proof. A calculation gives
fh () = eih f(),

(14.49)
so, by the Plancherel identity,

Z
(14.50)

kf

fh k2L2

|1 eih |2 |f()|2 d.

Now,

3
|h|
= |1 eih |2 2,
2
2

(14.51)
so

(14.52)

kf

fh k2L2

|f()|2 d.

3
2 |h| 2

Z
(14.53)
|f()|2 d Ch2 ,
2
4
h || h

Z
(14.51)

|f()|2 d C22` .

2` ||2`+1

Z
|f()| d
2` ||2`+1

n
(14.55)

|f()|2 d

2` ||2`+1

C2` 2`/2
= C2(1/2)` .

o1/2

1 d
2` ||2`+1

o1/2

123

Summing over ` N and using (again by Cauchys inequality)

Z
(14.56)
|f| d CkfkL2 = Ckf kL2 ,
||2

then gives the proof.

To see how close to sharp Proposition 14.8 is, consider
f (x) = I (x) = 1

(14.57)

if 0 x 1,
otherwise.

We have, for 0 h 1,
kf fh k2L2 = 2h,

(14.58)

so (14.47) holds, with = 1/2. Since A(R) C(R), this function does not belong to
A(R), so the condition (14.48) is about as sharp as it could be.
Remark. Using

(14.59)

Z
|gh| dx sup |g|

|h| dx,

(14.60)

x

so, with
(14.61)

kf kBV = sup kh1 (f fh )kL1 ,

0<h1

kf kC r =

sup
xR,0<h1

hr |f (x) fh (x)|,

(14.62)

14B. Fourier uniqueness

The Fourier inversion formula established in Corollary 14.3 yields
(14.63)
f A(R), f = 0 = f = 0.
Similarly, Proposition 14.4 yields
(14.64)

f L2 (R), f = 0 = f = 0.

We call these Fourier uniqueness results. An extension of (14.63) is the following consequence of Proposition 14.2:
(14.65)
f L1 (R) C(R), f = 0 = f = 0.
Here, we advertize the following strengthening of (14.61).

124

Proposition 14.9. We have the implication

f L1 (R), f = 0 = f = 0.

(14.66)

We indicate a proof of this result, starting with the following variant of (14.26). If
f L1 (R) and also g L1 (R), then
Z
(14.67)

Z Z
1

f ()g() d =
f (x)eix g() dx d
2

Z
=
f (x)
g (x) dx,

where the second identity uses a change in the order of integration. Thus
Z
(14.68)

f L (R), f = 0 =
1

f (x)h(x) dx = 0,

for all h = g, g L1 (R). In particular, (14.64) holds for all h A(R), and so, by (14.25),
it holds for all h C02 (R). The implication
Z
(14.69)

f L (R),

125

15. Laplace transforms

Suppose we have a function f : R+ C that is integrable on [0, R] for all R < and
satisfies
Z
(15.1)
|f (t)|eat dt < , a > A,
0

for some A (, ). We define the Laplace transform of f by

Z
(15.2)

f (t)est dt,

Lf (s) =

Re s > A.

It is clear that this integral is absolutely convergent for each s in the half-plane HA = {z
C : Re z > A} and defines a continuous function Lf : HA C. Also, if is a closed curve
(e.g., the boundary of a rectangle) in HA , we can change order of integration to see that
Z
(15.3)

Z
f (t)est ds dt = 0.

Lf (s) ds =
0

Hence Moreras theorem implies Lf is holomorphic on HA . We have

d
Lf (s) = Lg(s), g(t) = tf (t).
ds
R
On the other hand, if f C 1 ([0, )) and 0 |f 0 (t)|eat dt < for all a > A, then we
can integrate by parts and get
(15.4)

(15.5)

(15.6)

(15.7)

(15.8)

126

(15.9)

with
(15.10)

p(s) = cn sn + cn1 sn1 + + c0 ,

q(s) = cn (a0 sn1 + + an1 ) + + c1 a0 .

(15.11)

Lf (s) =

Lg(s) + q(s)
,
p(s)

and we are motivated to seek an inverse Laplace transform.

We can get this by relating the Laplace transform to the Fourier transform. In fact, if
(15.1) holds, and if B > A, then

(15.12)
Lf (B + i) = 2 (),

R,
with
(15.13)

(x) = f (x)eBx ,
0

x 0,
x < 0.

In 14 we have seen several senses in which

Z
1
ix
(15.14)
(x) =
()e

d,
2
hence giving, for t > 0,

(15.15)

Z
eBt
f (t) =
Lf (B + i)eit d
2
Z
1
=
Lf (s)est ds,
2i

where is the vertical line () = B + i, < < .

For example, if in (15.13) belongs to L2 (R), then (15.15) holds in the sense of Proposition 14.5. If belongs to A(R), then (15.15) holds in the sense of Corollary 14.3.
Frequently, f is continuous on [0, ) but f (0) 6= 0. Then in (15.13) has a discontinuity
at x = 0, so
/ A(R). However, sometimes one has (x) = x(x) in A(R), which is
obtained as in (15.13) by replacing f (t) by tf (t). In light of (15.4), we obtain
Z
1
d
(15.15A)
tf (t) =
Lf (s) est ds,
2i
ds

with an absolutely convergent integral, provided A(R).

Related to such inversion formulas is the following uniqueness result, which, via (15.12)
(15.13), is an immediate consequence of Proposition 14.9.

127

Proposition 15.1. If f1 and f2 are integrable on [0, R] for all R < and satisfy (15.1),
then
(15.16)

Lf1 (s) = Lf2 (s),

s HA = f1 = f2 on R+ .

We can also use material of 10 to deduce that f1 = f2 given Lf1 (s) = Lf2 (s) on a set
with an accumulation point in HA .

Exercises
1. Show that the Laplace transform of f (t) = tz1 , for Re z > 0, is given by
Lf (s) = (z) sz ,

(15.17)
where (z) is the Gamma function:

Z
(15.18)

(z) =

et tz1 dt.

For more on this function, see 18.

2. Compute the Laplace transforms of the following functions (defined for t 0).
(a)

eat ,

(b)

cosh at,

(c)

sinh at,

(d)

sin at,

(e)

tz1 eat .

3. Compute the inverse Laplace transforms of the following functions (defined in appropriate right half-spaces).
(a)
(b)
(c)
(d)
(e)

1
,
sa
s
,
2
s a2
a
,
2
s a2
a
,
2
s + a2
1

.
s+1

128

Reconsider these problems when you read 16.

Exercises 46 deal with the convolution of functions on R+ , defined by
Z
(15.19)

f g(t) =

f ( )g(t ) d.
0

4. Show that (15.19) coincides with the definition (14.44) of convolution, provided f (t)
and g(t) vanish for t < 0.
5. Show that if fz (t) = tz1 for t > 0, with Re z > 0, and if also Re > 0, then
(15.20)

with

Z
B(z, ) =

6. Show that
(15.21)

129

16. Residue calculus

Let f be holomorphic on an open set except for isolated singularities, at points pj .
Each pj is contained in a disk Dj on a neighborhood of which f has a Laurent series
(16.1)

f (z) =

an (pj )(z pj )n .

n=

We have
Z
1
(16.2)
Respj (f ) =
f (z) dz.
2i
Dj

If in addition is bounded, with piecewise smooth boundary, and f C( \ {pj }),

assuming {pj } is a finite set, we have, by the Cauchy integral theorem,
Z
(16.3)

f (z) dz =

X Z

f (z) dz = 2i

j D
j

Respj (f ).

This identity provides a useful tool for computing a number of interesting integrals of
functions whose anti-derivatives we cannot write down. Examples almost always combine
the identity (16.3) with further limiting arguments. We illustrate this method of residue
calculus to compute integrals with a variety of examples.
To start with a simple case, let us compute
Z

(16.4)

dx
.
1 + x2

In this case we can actually write down an anti-derivative, but never mind. The function
f (z) = (1 + z 2 )1 has simple poles at z = i. Whenever a meromorphic function f (z) has
a simple pole at z = p, we have
(16.5)

Resp (f ) = lim (z p)f (z).

zp

In particular,
(16.6)

Resi (1 + z 2 )1 = lim (z i)
zi

1
1
= .
(z + i)(z i)
2i

130

Let R be the path formed by the path R from R to R on the real line, followed by the
semicircle R of radius R in the upper half-plane, running from z = R to z = R. Then
R encloses the pole of (1 + z 2 )1 at z = i, and we have
Z

dz
= 2i Resi (1 + z 2 )1 = ,
1 + z2

(16.7)
R

provided R > 1. On the other hand, considering the length of R and the size of (1 + z 2 )1
on R , it is clear that
Z
(16.8)

lim

R
R

dz
= 0.
1 + z2

Hence
Z

(16.9)

dx
= lim
R
1 + x2

dz
= .
1 + z2

This is consistent with the result one gets upon recalling that d tan1 x/dx = (1 + x2 )1 .
For a more elaborate example, consider
Z

(16.10)

dx
.
1 + x4

(16.11)

p1 = ei/4 ,

p2 = e3i/4 ,

p3 = e3i/4 ,

p4 = ei/4 .

A computation using (16.5) gives

1 3i/4
e
,
4
1
= ei/4 .
4

Resp1 (1 + z 4 )1 =

(16.12)

Resp2 (1 + z 4 )1

Using the family of paths R as in (16.7), we have

Z
(16.13)

dx
= lim
R
1 + x4

2
X
dz
= 2i
Respj (1 + z 4 )1
1 + z4
j=1

i 3i/4
(e
+ ei/4 ) = ,
2
2

131

1+i
ei/4 = .
2

(16.14)

The evaluation of Fourier transforms provides a rich source of examples to which to

apply residue calculus. Consider the problem of computing
Z

(16.15)

eix
dx,
1 + x2

for R. Note that f (z) = eiz /(1 + z 2 ) has simple poles at z = i, and
(16.16)

eiz
e
=
.
1 + z2
2i

Resi

Hence, making use of R as in (16.7)(16.9), we have

Z

(16.17)

eix
dx = e ,
1 + x2

0.

For < 0 one can make a similar argument, replacing R by its image R reflected across
the real axis. Alternatively, we can see directly that (16.15) defines an even function of ,
so
Z
eix
(16.18)
dx = e|| , R.
2
1
+
x

The reader can verify that this result is consistent with the computation of
Z

e|x| eix dx

and the Fourier inversion formula; cf. Exercise 4 in 14.

As another example of a Fourier transform, we look at
Z
(16.19)

A=

eix
2 cosh

x
2

dx.

To evaluate this, we compare it with the integral over the path (x) = x 2i. See
Fig. 16.1. We have
Z
(16.20)

eiz
dz =
2 cosh z2

e2+ix
dx = e2 A,
2 cosh x2

132

since cosh(y i) = cosh y. Now the integrand has a pole at z = i, and a computation
gives
(16.21)

Resi

eiz
= i e .
2 cosh z2

We see that
A e2 A = 2i Resi

(16.22)

eiz
,
2 cosh z2

and hence
Z

(16.23)

eix

.
x dx =
2 cosh 2
cosh

The evaluation of (16.19) involved an extra wrinkle compared to the other examples
given above, involving how the integrand on one path is related to the integrand on another
path. Here is another example of this sort. Given (0, 1), consider the evaluation of
Z
(16.24)

B=
0

x
dx.
1 + x2

Let us define z = r ei upon writing z = rei , 0 2. Thus in contrast with our

treatment in 4, we here define z to be holomorphic on C \ R+ . It has distinct boundary
values as z = x + iy x > 0 as y & 0 and as y % 0. Let R be the curve starting with r
going from 0 to R, while = 0, then keeping r = R and taking from 0 to 2, and finally
keeping = 2 and taking r from R to 0. See Fig. 16.2. We have
Z
(16.25)

lim

R
R

z
dz = B e2i B.
1 + z2

On the other hand, for R > 1 we have

Z
(16.26)

X
z
z
dz
=
2i
Res
p
1 + z2
1 + z2
p=i

= i ei(1)/2 + e3i(1)/2 ,

so
(16.27)

B = i

ei(1)/2 + e3i(1)/2
cos (1 )/2
=
.
2i
1e
sin (1 )

133

While all of the examples above involved integrals over infinite intervals, one can also
use residue calculus to evaluate integrals of the form
Z

(16.28)

R(cos , sin ) d,
0

when R(u, v) is a rational function of its arguments. Indeed, if we consider the curve
() = ei , 0 2, we see that (16.28) is equal to
Z
1 z
1 dz
z
+ ,
,
(16.29)
R
2 2z 2i 2iz iz

which can be evaluated by residue calculus.

Exercises
1. Use residue calculus to evaluate the following definite integrals.
Z
x2
(a)
dx,
6
1 + x
Z

cos x
dx,
1 + x2

x1/3
dx,
1 + x3

(b)
0

Z
(c)
0

(d)
0

sin2 x
dx.
2 + cos x

2. Let R go from 0 to R on R+ , from R to Re2i/n on {z : |z| = R}, and from Re2i/n to

0 on a ray. Assume n > 1. Take R and evaluate
Z
dx
.
1 + xn
0
More generally, evaluate

Z
0

xa
dx,
1 + xn

134

3. If

n
k

Z
1
n
(1 + z)n
=
dz,
k
2i
z k+1

where is any simple closed curve about the origin.

Hint. nk is the coefficient of z k in (1 + z)n .
4. Use residue calculus to compute the inverse Laplace transforms in Exercise 3 (parts
(a)(d)) of 15.
5. Use the method involving (16.28)(16.29) to compute
Z

d
,
1 2r cos + r2

given 0 < r < 1. Compare your result with (13.8)(13.9).

6. Let f and g be holomorphic in a neighborhood of p.
(a) If g has a simple zero at p, show that
Resp

f
f (p)
= 0 .
g
g (p)

(16.30)

Resp

f (z)
= f 0 (p).
(z p)2

7. In the setting of Exercise 6, show that, for k 1,

(16.31)

Resp

f (z)
1
=
f (k1) (p).
k
(z p)
(k 1)!

Hint. Use (5.11), in concert with (16.2). See how this generalizes (16.5) and (16.30).
8. Compute

for R.

eix
dx,
1 + x4

135

17. The argument principle

Suppose C is a bounded domain with piecewise smooth boundary and f C 1 ()
is holomorphic on , and nowhere zero on . We desire to express the number of zeros
of f in in terms of the behavior of f on . We count zeros with multiplicity, where we
say pj is a zero of multiplicity k provided f (`) (pj ) = 0 for ` k 1 while f (k) (pj ) 6= 0.
The following consequence of Cauchys integral theorem gets us started.
Proposition 17.1. Under the hypotheses stated above, the number (f, ) of zeros of f
in , counted with multiplicity, is given by
Z 0
1
f (z)
(17.1)
(f, ) =
dz.
2i
f (z)

Proof. Suppose the zeros of f in occur at pj , with multiplicity mj , 1 j K. By

Cauchys integral theorem the right side of (17.1) is equal to
K

(17.2)

1 X
2i j=1

f 0 (z)
dz,
f (z)

Dj

(17.3)

1
mj =
2i

f 0 (z)
dz.
f (z)

Dj

(17.4)

(17.5)

gj0 (z)
f 0 (z)
mj
=
+
.
f (z)
z pj
gj (z)

The second term on the right is holomorphic on Dj , so it integrates to 0 over Dj . Hence

the identity (17.3) is a consequence of the known result
Z
dz
mj
.
(17.6)
mj =
2i
z pj
Dj

136

Having Proposition 17.1, we now want to interpret (17.1) in terms of winding numbers.
Denote the connected components of by Cj (with proper orientations). Say Cj is
parametrized by j : S 1 C. Then
f j : S 1 C \ 0

(17.7)

parametrizes the image curve j = f (Cj ), and we have

Z 0
Z
1
1
f (z)
dz
(17.8)
dz =
.
2i
f (z)
2i
z
j

Cj

Note that the right side of (17.1) is equal to the sum of the terms (17.8). As z runs over
j , we write z in polar coordinates, z = rei . Hence
(17.9)

dz = ei dr + irei d,

so
dr
dz
=
+ i d.
z
r
R
Noting that dr/r = d(log r) and that j d(log r) = 0, we have
Z
Z
1
1
dz
(17.11)
=
d.
2i
z
2
(17.10)

This theta integral is not necessarily zero, since is not single valued on C \ 0. Rather we
have, for any closed curve in C \ 0,
Z
1
(17.12)
d = n(, 0),
2

the winding number of about 0. Certainly the amount by which changes over such a
curve is an integral multiple of 2, so n(, 0) is an integer.
We can provide a formula for d in terms of single valued functions, as follows. We start
with

(17.13)

dz
dx + idy
(x iy)(dx + idy)
=
=
z
x + iy
x2 + y 2
x dx + y dy
x dy y dx
=
+i 2
,
2
2
x +y
x + y2

(17.14)

d =

x dy y dx
.
x2 + y 2

137

(17.15)

Z
1
(17.16)
n(s , 0) =
d
2
s

is a continuous function of s [0, 1], taking values in Z. Hence it is constant.

Comparing (17.8), (17.11), and (17.12), we have
Proposition 17.3. With the winding number given by (17.12),
(17.17)

1
2i

f 0 (z)
dz = n(j , 0),
f (z)

j = f (Cj ).

Cj

In concert with Proposition 17.1, this yields:

Proposition 17.4. In the setting of Proposition 17.1, with Cj denoting the connected
components of ,
(17.18)

(f, ) =

n(j , 0),

j = f (Cj ).

That is, the total number of zeros of f in , counting multiplicity, is equal to the sum of
the winding numbers of f (Cj ) about 0.
This result is called the argument principle. It is of frequent use, since the right side
of (17.18) is often more readily calculable directly than the left side. In evaluating this
sum, take care as to the orientation of each component Cj , as that affects the sign of the
winding number. We mention without further ado that the smoothness hypothesis on
can be relaxed via limiting arguments.
The following useful corollary of Proposition 17.4 is known as Rouches theorem or the
dog-walking theorem.
Proposition 17.5. Let f, g C 1 () be holomorphic in , and nowhere zero on .
Assume
(17.19)

|f (z) g(z)| < |f (z)|,

z .

Then
(17.20)

(f, ) = (g, ).

138

Proof. The hypothesis (17.19) implies that f and g are smoothly homotopic as maps from
to C \ 0, e.g., via the homotopy
f (z) = f (z) [f (z) g(z)],

0 1.

Hence, by Proposition 17.2, f |Cj and g|Cj have the same winding numbers about 0, for
each boundary component Cj .
As an example of how this applies, we can give another proof of the fundamental theorem
of algebra. Consider
(17.21)

f (z) = z n ,

g(z) = z n + an1 z n1 + + a0 .

(17.22)

for |z| = R.

(17.23)

so the polynomial g(z) has complex roots.

The next corollary of Proposition 17.4 is known as the open mapping theorem for
holomorphic functions.
Proposition 17.6. If C is open and connected and f : C is holomorphic and
non-constant, then f maps open sets to open sets.
Proof. Suppose p and q = f (p). We have a power series expansion
(17.24)

f (z) = f (p) +

an (z p)n ,

n=k

where we pick ak to be the first nonzero coefficient. It follows that there is a disk D (p)
such that = f D (p) is bounded away from q, and the winding number of this curve

about q is equal to k (which is 1). It follows that there exists > 0 such that whenever
|q 0 q| < then the winding number of about q 0 is equal to k, so f (z) q 0 has zeros
in D (p). This shows that {q 0 C : |q 0 q| < } is contained in the range of f , so the
proposition is proven.
The argument principle also holds for meromorphic functions. We have the following
result.
Proposition 17.7. Assume f is meromorphic on a bounded domain , and C 1 on a
neighborhood of . Then the number of zeros of f minus the number of poles of f
(counting multiplicity) in is equal to the sum of the winding numbers of f (Cj ) about 0,
where Cj are the connected components of .
Proof. The identity (17.1), with (f, ) equal to zeros minus poles, follows by the same
reasoning as used in the proof of Proposition 17.1, and the interpretation of the right side
of (17.1) in terms of winding numbers follows as before.
Another application of Proposition 17.1 yields the following result, known as Hurwitz
theorem.

139

Proposition 17.8. Assume fn are holomorphic on a connected region and fn f

locally uniformly on . Assume each fn is nowhere vanishing in . Then f is either
nowhere vanishing or identically zero in .
Proof. We know f is holomorphic in and fn0 f 0 locally uniformly on ; see Exercise 1
of 5. Assume f is not identically zero. If it has zeros in , they are isolated. Say D is a
disk in such that f has zeros in D but not in D. It follows that 1/fn 1/f uniformly
on D. By (17.1),
1
2i

(17.25)

Z
D

fn0 (z)
dz = 0,
fn (z)

n.

(17.26)

1
(f, D) =
2i

f 0 (z)
dz = 0,
f (z)

contradicting the possibility that f has zeros in D.

Exercises
1. Let f (z) = z 3 + iz 2 2iz + 2. Compute the change in the argument of f (z) as z varies
along:
a) the real axis from 0 to ,
b) the imaginary axis from 0 to ,
c) the quarter circle z = Rei , where R is large and 0 /2.
Use this information to determine the number of zeros of f in the first quadrant.
2. Prove that for any > 0 the function
1
+ sin z
z+i
has infinitely many zeros in the strip | Im z| < .
Hint. Rouches theorem.
3. Suppose f : C is holomorphic and one-to-one. Show that f 0 (p) 6= 0 for all p ,
using the argument principle.
Hint. Compare the proof of Proposition 17.6, the open mapping theorem.
4. Make use of Exercise 7 in 5 to produce another proof of the open mapping theorem.

140

5. Let C be open and connected and assume gn : C are holomorphic and each
is one-to one (we say univalent). Assume gn g locally uniformly on . Show that g is
either univalent or constant.
Hint. Pick arbitrary b and consider fn (z) = gn (z) gn (b).
6. Let D be a disk in C. Assume f C 1 (D) is holomorphic in D. Show that f (D)
cannot be a figure 8.
7. In the setting of Proposition 17.1, assume S C is connected and S f () = . Show
that
(f p, ) is independent of p S.
Hint. Use (17.1) to show that (p) = (f p, ) gives a continuous function : S Z.
8. Let > 1. Show that zez = 1 for exactly one z D = {z C : |z| < 1}.
Hint. With f (z) = zez , show that
|z| = 1 = |f (z)| > 1.
Compare the number of solutions to f (z) = 0. Use either Exercise 17, with S = D, or
Rouches theorem. with f (z) = zez and g(z) = zez 1.
In Exercises 912, we consider
(z) = ze

: D C,

: D C.

9. Show that : D C is one-to-one.

Hint. (z) = (w) z/w = ezw z w iR z = w
(z) = (z) ei(sin ) = 1 if z = ei
Given Exercise 9, it is a consequence of the Jordan curve theorem (which we assume here)
that C \ (D) has exactly two connected components. Say + is the component that
contains 0 and is the other one.
10. Show that
p + ( p, D) = 1,

p ( p, D) = 0.

Hint. For the first case, take p = 0. For the second, let p .
11. Deduce that : D + is one-to-one and onto.
12. Recalling Exercise 8, show that {z C : |z| < 1/e} + .

141

18. The Gamma function

The Gamma function has been previewed in (15.17)(15.18), arising in the computation
of a natural Laplace transform:
(18.1)

for Re z > 0, with

(18.2)

(z) =

et tz1 dt,

Re z > 0.

Here we develop further properties of this special function, beginning with the following
crucial identity:
Z
(z + 1) =
et tz dt
0
Z
d t z
(18.3)
(e ) t dt
=
dt
0
= z (z),
for Re z > 0, where we use integration by parts. The definition (18.2) clearly gives
(18.4)

(1) = 1,

(18.5)

(k) = (k 1)(k 1) = = (k 1)!.

While (z) is defined in (18.2) for Re z > 0, note that the left side of (18.3) is well defined
for Re z > 1, so this identity extends (z) to be meromorphic on {z : Re z > 1}, with
a simple pole at z = 0. Iterating this argument, we extend (z) to be meromorphic on C,
with simple poles at z = 0, 1, 2, . . . . Having such a meromorphic continuation of (z),
we establish the following identity.
Proposition 18.1. For z C \ Z we have
(18.6)

(z)(1 z) =

.
sin z

Proof. It suffices to establish this identity for 0 < Re z < 1. In that case we have
Z Z
(z)(1 z) =
e(s+t) sz tz1 ds dt
Z0 Z0
(18.7)
=
eu v z1 (1 + v)1 du dv
Z0 0
=
(1 + v)1 v z1 dv,
0

142

where we have used the change of variables u = s + t, v = t/s. With v = ex , the last
integral is equal to
Z
(18.8)
(1 + ex )1 exz dx,

which is holomorphic on 0 < Re z < 1. We want to show that this is equal to the right
side of (18.6) on this strip. It suffices to prove identity on the line z = 1/2 + i, R.
Then (18.8) is equal to the Fourier integral
Z
x 1 ix
(18.9)
2 cosh
e dx.
2

,
cosh

(18.10)
and since

(18.11)

sin ( 12 + i)

,
cosh

the demonstration of (18.6) is complete.

Corollary 18.2. The function (z) has no zeros, so 1/(z) is an entire function.
For our next result, we begin with the following estimate:
Lemma 18.3. We have
(18.12)

t n
t2
0 et 1
et ,
n
n

0 t n,

the latter inequality holding provided n 4.

Proof. The first inequality in (18.12) is equivalent to the simple estimate ey (1 y) 0
for 0 y 1. To see this, denote the function by f (y) and note that f (0) = 0 while
f 0 (y) = 1 ey 0 for y 0.
As for the second inequality in (18.12), write

(18.13)

t n
t
log 1
= n log 1
= t X,
n
n

1 t 2
t2 1 1 t
X=
+
+
+ .
n 2 3n 4 n

We have (1 t/n)n = etX and hence, for 0 t n,

t n
et 1
= (1 eX )et Xet ,
n

143

using the estimate x (1 ex ) 0 for x 0 (as above). It is clear from (18.13) that
X t2 /n if t n/2. On the other hand, if t n/2 and n 4 we have t2 /n 1 and hence
et (t2 /n)et .
We use (18.12) to obtain, for Re z > 0,
Z n
t n z1
(z) = lim
1
t
dt
n 0
n
Z 1
z
= lim n
(1 s)n sz1 ds.
n

Repeatedly integrating by parts gives

(18.14)

n(n 1) 1
(z) = lim n
n
z(z + 1) (z + n 1)

sz+n1 ds,

which yields the following result of Euler:

Proposition 18.4. For Re z > 0, we have
(z) = lim nz

(18.15)

1 2n
,
z(z + 1) (z + n)

Using the identity (18.3), analytically continuing (z), we have (18.15) for all z C
other than 0, 1, 2, . . . . In more detail, we have
(z) =

(z + 1)
1
1 2n
= lim nz+1
,
n
z
z (z + 1)(z + 2) (z + 1 + n)

for Re z > 1(z 6= 0). We can rewrite the right side as

1 2n n
z(z + 1) (z + n + 1)
n z+1
1 2 (n + 1)
= (n + 1)z

,
z(z + 1) (z + n + 1)
n+1

nz

and (n/(n + 1))z+1 1 as n . This extends (18.15) to {z 6= 0 : Re z > 1}, and

iteratively we get further extensions.
We can rewrite (18.15) as

z 1
z 1
z 1
1
(18.16)
(z) = lim n z (1 + z)
1+
1 +
.
n
2
n
To work on this formula, we define Eulers constant:

1
1
(18.17)
= lim 1 + + + log n .
n
2
n
Then (18.16) is equivalent to

z 1
z 1
z z(1+1/2++1/n) 1
1
1+
(18.18)
(z) = lim e
e
z (1 + z)
1 +
,
n
2
n
which leads to the following Euler product expansion.

144

Proposition 18.5. For all z C, we have

Y
1
z z/n
= z ez
1+
e
.
(z)
n
n=1

(18.19)

We can combine (18.6) and (18.19) to produce a product expansion for sin z. In fact,
it follows from (18.19) that the entire function 1/(z)(z) has the product expansion

Y
z2
1
2
= z
1 2 .
(z)(z)
n
n=1

(18.20)

(18.21)

sin z = z

Y
n=1

z2
1 2 .
n

For another proof of this result, see 30, Exercise 2.

Here is another application of (18.6). If we take z = 1/2, we get (1/2)2 = . Since
(18.2) implies (1/2) > 0, we have
1
(18.22)

= .
2
Another way to obtain (18.22) is the following. A change of variable gives
Z
Z
1 t 1/2
1 1
x2
(18.23)
e
dx =
e t
dt =
.
2 0
2 2
0

It follows from (10.6) that the left side of (18.23) is equal to /2, so we again obtain
(18.22). Note that application of (18.3) then gives, for each integer k 1,

1
1
1
3
(18.24)
k+
= 1/2 k
k

.
2
2
2
2
One can calculate the area An1 of the unit sphere S n1 Rn by relating Gaussian
integrals to the Gamma function. To see this, note that the argument giving (10.6) yields
Z
Z
n
2
|x|2
e
dx =
ex dx = n/2 .
(18.25)
Rn

On the other hand, using spherical polar coordinates to compute the left side of (18.24)
gives
Z
Z
2
|x|2
e
dx = An1
er rn1 dr
(18.26)

Rn

1
= An1
2

Z
0

et tn/21 dt,

145

(18.27)

An1 =

2 n/2
.
(n/2)

More details on this argument are given at the end of Appendix C.

Exercises
1. Use the product expansion (18.19) to prove that

d 0 (z) X
1
=
.
2
dz (z)
(z
+
n)
n=0

(18.28)
Hint. Go from (18.19) to

X
1
z zi
log
= log z + z +
log 1 +

,
(z)
n
n
n=1

and note that

d 0 (z)
d2
= 2 log (z).
dz (z)
dz

2. Let

1
1
+ + log(n + 1).
2
n
Show that n % and that 0 < n < 1. Deduce that = limn n exists, as asserted in
(18.17).
n = 1 +

3. Using (/z)tz1 = tz1 log t, show that

fz (t) = tz1 log t,

(Re z > 0)

Lfz (s) =

,
sz

Re s > 0.

4. Show that (18.19) yields

(18.29)

(z + 1) = z(z) = e

Y
z 1 z/n
1+
e ,
n
n=1

|z| < 1.

146

0 (1) =

(18.30)

d
z(z) z=0 = .
dz

5. Using Exercises 34, show that

f (t) = log t = Lf (s) =
and that

log s +
,
s

Z

a =
0

1 et
dt,
t

b =
1

et
dt.
t

Consider how to obtain accurate numerical evaluations of these quantities.

Hint. Split the integral for in Exercise 5 into two pieces. Integrate each piece by parts,
using et = (d/dt)(et 1) for one and et = (d/dt)et for the other. See Appendix
J for more on this.
7. Use the Laplace transform identity (18.1) for fz (t) = tz1 (on t 0, given Re z > 0)
plus the results of Exercises 56 of 15 to show that
(18.31)

B(z, ) =

(z)()
,
(z + )

Re z, Re > 0,

Z
(18.32)

B(z, ) =

sz1 (1 s)1 ds,

Re z, Re > 0.

The identity (18.31) is known as Eulers formula for the beta function.
8. Show that, for any z C, when n 2|z|, we have

z z/n
e
= 1 + wn
1+
n
with log(1 + wn ) = log(1 + z/n) z/n satisfying
2

log(1 + wn ) |z| .
n2

147

Show that this estimate implies the convergence of the product on the right side of (18.19),
locally uniformly on C.
More infinite products
9. Show that

Y

(18.33)

n=1

1
2
1 2 = .
4n

Hint. Take z = 1/2 in (18.21).

10. Show that, for all z C,
(18.34)

cos

z
=
2

Y
z2
1 2 .
n

odd n1

Y
z

(z 1)2
cos
= (1 z)
1
.
2
2
2
4n
n=1

(18.35)

1+

1
z
1
z

1
+
2n 2n
2n 2n

1
z
z
= 1 2 1
1+
,
4n
2n + 1
2n 1

z
Y
1
cos
=
1 2
2
2 n=1
4n

Y
odd n1

z
z
1
1+
.
n
n

11. Show that
(18.36)

sin z
z
z
z
= cos
cos
cos
.
z
2
4
8

148

18A. The Legendre duplication formula

The Legendre duplication formula relates (2z) and (z)(z + 1/2). Note that each
of these functions is meromorphic, with poles precisely at {0, 1/2, 1, 3/2, 2, . . . },
all simple, and both functions are nowhere vanishing. Hence their quotient is an entire
holomorphic function, and it is nowhere vanishing, so

1
(18.37)
(2z) = eA(z) (z) z +
,
2
with A(z) holomorphic on C. We seek a formula for A(z). We will be guided by (18.19),
which implies that

Y
2z 2z/n
1
2z
= 2ze
1+
e
,
(2z)
n
n=1

(18.38)

and (via results given in 18B)

(18.39)

1
(z)(z + 1/2)

1 z (z+1/2) n Y
z z/n
z + 1/2 (z+1/2)/n o
=z z+
e e
1+
e
1+
e
.
2
n
n
n=1

Setting
(18.40)

2z + 2n + 1
2z
1
z + 1/2
=
= 1+
1+
,
1+
n
2n
2n + 1
2n

and
(18.41)

we can write the infinite product on the right side of (18.39) as

nY

(18.42)

1+

n=1

2z 2z/2n
2z 2z/(2n+1) o
e
1+
e
2n
2n + 1

nY
n=1

1+

1 1/2n o Y 2z[(1/2n)1/(2n+1)]
e

e
.
2n
n=1

Hence

(18.43)

1
e2z
= ze2z e/2
(1 + 2z)e2z (18.42)
(z)(z + 1/2)
2

2z n Y
2z 2z/k o
2z /2 e
= 2ze e
1+
e
4
k
k=1

nY
n=1

1 1/2n o Y 2z[(1/2n)1/(2n+1)]
1+
e
e
.
2n
n=1

149

1
1 /2 Y
1 1/2n
= e
1+
e
,
(1/2)
2
2n
n=1

(18.44)

so taking (18.38) into account yields

(18.45)

1
1
e2z Y 2z[(1/2n)1/(2n+1)]
=
e
(z)(z + 1/2)
(1/2)(2z) 2 n=1

e2z
1
,
(1/2)(2z) 2

where

(18.46)

X
1
1
=1

2n 2n + 1
n=1

1 1 1 1
+ +
2 3 4 5
= log 2.

=1

1

1
(18.47)

(2z) = 22z1 (z) z +

.
2
2
This is the Legendre duplication formula. Recall that (1/2) =
An equivalent formulation of (18.47) is
z z + 1
1/2
z1/2
(18.48)
(2) (z) = 2

.
2
2

This generalizes to the following formula of Gauss,

z z + 1
z + n 1
(18.49)
(2)(n1)/2 (z) = nz1/2

,
n
n
n
valid for n = 3, 4, . . . .

18B. Convergence of infinite products

Here we record some results regarding the convergence of infinite products, which have
arisen in this section. We look at infinite products of the form
(18.50)

(1 + ak ).

k=1

150

Disregarding cases where one or more factors 1+ak vanish, the convergence of
as M amounts to the convergence
(18.51)

lim

N
Y

(1 + ak ) = 1,

QM

k=1 (1+ak )

uniformly in N > M.

k=M

In particular, we require ak 0 as k . To investigate when (18.51) happens, write

N
Y

(18.52)

(1 + ak ) = (1 + aM )(1 + aM +1 ) (1 + aN )

k=M

=1+

aj +

aj1 aj2 + + aM aN ,

j1 <j2

where, e.g., M j1 < j2 N . Hence

N
X
Y
X

|aj | +
|aj1 aj2 | + + |aM aN |
(1
+
a
)

k
j

k=M

(18.53)
=

j1 <j2

N
Y

(1 + |ak |) 1

k=M

= bM N ,
the last identity defining bM N . Our task is to investigate when bM N 0 as M ,
uniformly in N > M . To do this, we note that
N
Y

log(1 + bM N ) = log

(1 + |ak |)

k=M

(18.54)
=

N
X

k=M

and use the facts

(18.55)

x 0 = log(1 + x) x,
x
0 x 1 = log(1 + x) .
2

Assuming ak 0 and taking M so large that k M |ak | 1/2, we have

(18.56)

N
N
X
1 X
|ak | log(1 + bM N )
|ak |,
2
k=M

k=M

151

and hence
(18.57)

lim bM N = 0, uniformly in N > M

|ak | < .

Consequently,
X

|ak | < =

(18.58)

Y
k=1

(1 + |ak |) converges
(1 + ak ) converges.

k=1

(18.59)

If 1 + ak 6= 0 for all k, then

|ak | <

(1 + ak ) 6= 0.

k=1

We can replace the sequence (ak ) of complex numbers by a sequence (fk ) of holomorphic
functions, and deduce from the estimates above the following.
Proposition 18.6. Let fk : C be holomorphic. If
X

(18.60)

then we have a convergent infinite product

(18.61)

(1 + fk (z)) = g(z),

k=1

and g is holomorphic on . If z0 and 1 + fk (z0 ) 6= 0 for all k, then g(z0 ) 6= 0.

P Another consequence of estimates leading to (18.57) is that if also gk : C and
|gk (z)| < on , then
(18.62)

nY
k=1

oY
Y
(1 + fk (z))
(1 + gk (z)) =
(1 + fk (z))(1 + gk (z)).
k=1

k=1

To make contact with the Gamma function, note that the infinite product in (18.19)
has the form (18.61) with
(18.63)

z z/k
e
.
1 + fk (z) = 1 +
k

152

(18.64)

ew = 1 w + R(w),

Hence

1+
(18.65)

z
z z/k
z
z
e
= 1+
1 +R
k
k
k
k

2
z
z
z
R
.
=1 2 + 1
k
k
k

Hence (18.63) holds with

(18.66)

fk (z) =

z z
z2
+
1

R
,
k2
k
k

so
(18.67)
which yields (18.60).

z 2

|fk (z)| C for k |z|,
k

153

19. The Riemann zeta function

The Riemann zeta function is defined by
(19.1)

(s) =

ns ,

Re s > 1.

n=1

Some special cases of this arose in 13, namely

(19.2)

2
,
6

(2) =

(4) =

4
.
90

This function is of great interest in number theory, due to the following result.
Proposition 19.1. Let {pj : j 1} = {2, 3, 5, 7, 11, . . . } denote the set of prime numbers
in N. Then, for Re s > 1,
(19.3)

(s) =

1
(1 ps
.
j )

j=1

(19.4)

2s
(1 + ps
+ p3s
+ )
j + pj
j

j=1

=1+

ps
j +

(pj1 pj2 )s +

j1 j2

(pj1 pj2 pj3 )s + .

j1 j2 j3

That this is identical to the right side of (19.1) follows from the fundamental theorem of
arithmetic, which says that each integer n 2 has a unique factorization into a product
of primes.
From (19.1) we see that
(19.5)

s & 1 = (s) % +.

Hence
(19.6)

(1 p1
j ) = 0.

j=1

154

X
1
= ,
p
j=1 j

(19.7)

which is a quantitative strengthening

of the result that there are infinitely many primes.
P
Of course, comparison with n1 ns implies

X
1
s | < ,
|p
j
j=1

(19.8)

for Re s > 1.

(19.9)

(s) 6= 0,

for Re s > 1.

Our next goal is to establish the following.

Proposition 19.2. The function (s) extends to a meromorphic function on C, with one
simple pole, at s = 1.
To start the demonstration, we relate the Riemann zeta function to the function
(19.10)

g(t) =

en

n=1

Indeed, we have
Z
(19.11)

g(t)t

s1

dt =

2s s

n=1

= (2s)

et ts1 dt

(s).

This gives rise to further useful identities, via the Jacobi identity (14.43), i.e.,
r

X
1 X k2 /t
`2 t
(19.12)
e
=
e
,
t
`=

k=

which implies
(19.13)

1
1 1 1/2
1/2
g(t) = + t
+t
g
.
2 2
t

To use this, we first note from (19.11) that, for Re s > 1,

Z
s
s/2

(s) =
g(t)ts/21 dt
2
0
(19.14)
Z 1
Z
s/21
=
g(t)t
dt +
g(t)ts/21 dt.
0

155

Into the integral over [0, 1] we substitute the right side of (19.13) for g(t), to obtain
Z 1
s

1 1
s/2

(s) =
+ t1/2 ts/21 dt
2
2 2
0
(19.15)
Z 1
Z
g(t1 )ts/23/2 dt +

+
0

g(t)ts/21 dt.

We evaluate the first integral on the right, and replace t by 1/t in the second integral, to
obtain, for Re s > 1,
Z
s
s/2

1
1
s/2
(19.16)

(s) =
+
t + t(1s)/2 g(t)t1 dt.
2
s1 s
1
Note that g(t) Cet for t [1, ), so the integral on the right side of (19.16) defines
an entire function of s. Since 1/(s/2) is entire, with simple zeros at s = 0, 2, 4, . . . , as
seen in 18, this implies that (s) is continued as a meromorphic function on C, with one
simple pole, at s = 1. This finishes the proof of Proposition 19.2.
The formula (19.16) does more than establish the meromorphic continuation of the zeta
function. Note that the right side of (19.16) is invariant under replacing s by 1 s. Thus
we have an identity known as Riemanns functional equation:
s
1 s
(19.17)

s/2 (s) =
(1s)/2 (1 s).
2
2
The meromorphic continuation of (s) can be used to obtain facts about the set of
primes deeper than (19.7). It is possible to strengthen (19.9) to
(19.18)

(s) 6= 0

for Re s 1.

This plays a role in the following result, known as the

Prime number theorem.
(19.19)

lim

pj
= 1.
j log j

A proof can be found in [BN], and in [Ed]. Of course, (19.19) is much more precise than
(19.7).
There has been a great deal of work on determining where (s) can vanish. By (19.16),
it must vanish at all the poles of (s/2), other than s = 0, i.e.,
(19.20)

(s) = 0 on {2, 4, 6, . . . }.

These are known as the trivial zeros of (s). It follows from (19.18) and the functional
equation (19.16) that all the other zeros of (s) are contained in the critical strip
(19.21)

= {s C : 0 < Re s < 1}.

Concerning where in these zeros can be, there is the following famous conjecture.

156

The Riemann hypothesis. All zeros in of (s) lie on the critical line
n1

(19.22)

o
+ i : R .

Many zeros of (s) have been computed and shown to lie on this line, but after over a
century, a proof (or refutation) of the Riemann hypothesis eludes the mathematics community. The reader can consult [Ed] for more on the zeta function.

Exercises
1. Use the functional equation (19.17) together with (18.6) and the Legendre duplication
formula (18.47) to show that

s
(1 s) = 21s s cos
(s)(s).
2
2. Sum the identity

Z
s

(s)n

ent ts1 dt

over n Z+ to show that

Z
(s)(s) =
0

ts1
dt =
et 1

Z
0

ts1
dt +
et 1

Z
1

ts1
dt = A(s) + B(s).
et 1

Show that B(s) continues as an entire function. Use a Laurent series

et
to show that

1
0

1
1
= + a0 + a1 t + a2 t2 +
1
t

ts1
1
a0
a1
dt =
+
+
+
t
e 1
s1
s
s+1

provides a meromorphic continuation of A(s), with poles at {1, 0, 1, . . . }. Use this to give
a second proof that (s) has a meromorphic continuation with one simple pole, at s = 1.

157

3. Show that, for Re s > 1, the following identities hold:

(a)

X
d(n)
(s) =
,
ns
n=1

(b)

X
(n)
(s)(s 1) =
,
ns
n=1

(c)

(s 1) X (n)
=
,
s
(s)
n
n=1

(d)

X
1
(n)
=
,
(s) n=1 ns

(e)
where

X
0 (s)
(n)
=
,
s
(s)
n
n=1

d(n) = # divisors of n,
(n) = sum of divisors of n,
(n) = # positive integers n, relatively prime to n,
(n) = (1)# prime factors , if n is square-free, 0 otherwise,
(n) = log p if n = pm for some prime p, 0 otherwise.

158

20. Covering maps and inverse functions

The concept of covering map comes from topology. Generally, if E and X are topological
spaces, a continuous map : E X is said to be a covering map provided every p X
has a neighborhood U such that 1 (U ) is a disjoint union of open sets Sj E, each
of which is mapped homeomorphically by onto U . In topology one studies conditions
under which a continuous map f : Y X lifts to a continuous map f : Y E, so that
f = f. Here is one result, which holds when Y is simply connected. By definition, a
connected, locally path connected space Y is said to be simply connected provided that
whenever is a closed path in Y , there exists a continuous family s (0 s 1) of closed
paths in Y such that 1 = and the image of 0 consists of a single point. (Cf. 5.)
Proposition 20.1. Assume E, X, and Y are all connected and locally path-connected, and
: E X is a covering map. If Y is simply connected, any continuous map f : Y X
lifts to f : Y E.
A proof can be found in Chapter 6 of [Gr]. See also Chapter 8 of [Mun]. Here our
interest is in holomorphic covering maps : O, where and O are domains in C.
The following is a simple but useful result.
Proposition 20.2. Let U, and O be connected domains in C. Assume : O is
a holomorphic covering map and f : U O is holomorphic. Then any continuous lift
f : U of f is also holomorphic.
Proof. Take q U, p = f (q) O. Let V be a neighborhood of p such that 1 (V) is
a disjoint union of open sets Sj with : Sj V a (holomorphic) homeomorphism.
As we have seen (in several previous exercises) this implies : Sj V is actually a
holomorphic diffeomorphism.
Now f(q) Sk for some k, and f1 (Sk ) = f 1 (V) = Uq is a neighborhood of q in U .
We have

fU = 1 f U ,
q
q

(20.1)

which implies f is holomorphic on Uq , for each q U . This suffices.

Proposition 20.3. The following are holomorphic covering maps:
(20.2)

exp : C C \ {0},

where exp(z) = ez and Sq(z) = z 2 .

Proof. Exercise.

Sq : C \ {0} C \ {0},

159

Corollary 20.4. If U C is simply connected and f : U C \ {0} is holomorphic, then

there exist holomorphic functions g and h on U such that
(20.3)

f (z) = eg(z) ,

f (z) = h(z)2 .

We say g(z) is a branch of log f (z) and h(z) is a branch of

p
f (z), over U .

Exercises
1. As in Corollary 20.4, suppose U C is simply connected and f : U C \ {0} is
holomorphic.
(a) Show that f 0 /f : U C is holomorphic.
(b) Pick p U and define

Z
(z) =
p

f 0 ()
d.
f ()

Show this is independent of the choice of path in U from p to z, and it gives a holomorphic
function : U C.
(c) Use to give another proof of Corollary 20.4 (particularly the existence of g(z) in
(20.3)).

160

21. Normal families

Here we discuss a certain class of sets of mappings, the class of normal families. In a
general context, suppose is a locally compact metric space and S is a complete metric
space. Let C(, S) denote the set of continuous maps f : S. We say a subset
F C(, S) is a normal family (with respect to (, S)) if and only if the following
property holds:
(21.1)

Every sequence f F has a locally uniformly

convergent subsequence fk f C(, S).

If the identity of the pair (, S) is understood, we omit the phrase with respect to (, S).
The main case of interest to us here is where C is an open set and S = C. However,
in later sections the case S = C {} S 2 will also be of interest.
A major technique to identify normal families is the following result, known as the
Arzela-Ascoli theorem.
Proposition 21.1. Let X and Y be compact metric spaces and fix a modulus of continuity
(). Then
(21.2)

is a compact subset of C(X, Y ), hence a normal family.

This result is given as Proposition A.18 in Appendix A and proven there. See also this
appendix for a discussion of C(X, Y ) as a metric space. The defining condition
(21.3)

x, y X, f F ,

for some modulus of continuity is called equicontinuity of F. The following result is a

simple extension of Proposition 21.1.
Proposition 21.2. Assume there exists a countable family {Kj } of compact subsets of
such that any compact K is contained in some finite union of these Kj . Consider
a family F C(, S). Assume that, for each j, there exist compact Lj S such that
f : Kj Lj for all f F , and that {f |Kj : f F} is equicontinuous. Then F is a
normal family.
Proof. Let f be a sequence in F. By Proposition 21.1 there is a uniformly convergent
subsequence fk : K1 L1 . This has a further subsequence converging uniformly on K2 ,
etc. A diagonal argument finishes the proof.
The following result is sometimes called Montels theorem, though there is a deeper
result, discussed in 27, which is more properly called Montels theorem.

161

Proposition 21.3. Let C be open. A family F of holomorphic functions f : C

is normal (with respect to (, C)) if and only if this family is uniformly bounded on each
compact subset of .
Proof. We can write = j Dj for a countable family of closed disks Dj , and satisfy
the hypothesis on in Proposition 21.2. Say dist(z, ) j > 0 for all z Dj . The
hypothesis
(21.4)

|f | Aj on Dj ,

f F,

plus Cauchys estimate (5.32) gives

(21.5)

|f0 |

Aj
on Dj ,
j

f F ,

hence
(21.6)

|f (z) f (w)|

Aj
|z w|,
j

z, w Dj , f F.

This equicontinuity on each Dj makes Proposition 21.2 applicable. This establishes one
implication in Proposition 21.3, and the reverse implication is easy.

Exercises
1. Show whether each of the following families is or is not normal (with respect to (, C)).
(a) {n1 cos nz : n = 1, 2, 3, . . . },

(b) The set of holomorphic maps g : D U such that g(0) = 0, with

= D = {z : |z| < 1},

2. Suppose that F is a normal family (with respect to (, C)). Show that {f 0 : f F} is

also a normal family. (Compare Exercise 1 in 26.)
3. Let F be the set of entire functions f such that f 0 (z) has a zero of order one at z = 3
and satisfies
|f 0 (z)| 5|z 3|, z C.
Find all functions in F. Determine whether F is normal (with respect to (C, C)).
4. Let F = {z n : n Z+ }. For which regions is F normal with respect to (, C)?
(Compare Exercise 3 in 26.)

162

22. Conformal maps

In this section we explore geometrical properties of holomorphic diffeomorphisms f :
O between various domains in C. These maps are also called biholomorphic maps,
and they are also called conformal maps. Let us explain the latter terminology.
A diffeomorphism f : O between two planar domains is said to be conformal
provided it preserves angles. That is, if two curves 1 and 2 in meet at an angle at p,
then 1 = f (1 ) and 2 = f (2 ) meet at the same angle at q = f (p). The condition for
this is that, for each p , Df (p) End(R2 ) is a positive multiple (p) of an orthogonal
matrix:
(22.1)

Df (p) = (p)R(p).

Now det Df (p) > 0 det R(p) = +1 and det Df (p) < 0 det R(p) = 1. In the
former case, R(p) has the form

(22.2)

R(p) =

cos
sin

sin
cos

of a rotation matrix, and we see that Df (p) commutes with J, given by (1.39). In the
latter case, R(p) has the form

(22.3)

R(p) =

cos
sin

sin
cos

(22.4)

C=

1
0

0
1

i.e., Cz = z. We have the following result.

Proposition 22.1. Given planar regions , O, the class of orientation-preserving conformal diffeomorphisms f : O coincides with the class of holomorphic diffeomorphisms.
The class of orientation-reversing conformal diffeomorphisms f : O coincides with
the class of conjugate-holomorphic diffeomorphisms.
There are some particular varieties of conformal maps that have striking properties.
Among them we first single out the linear fractional transformations. Given an invertible
2 2 complex matrix A (we say A Gl(2, C)), set
(22.5)

az + b
LA (z) =
,
cz + d

A=

a b
c d

163

(22.6)

LA : C {} C {}

by setting
(22.7)

LA (d/c) = ,

if c 6= 0,

and
(22.8)

LA () =

a
if c 6= 0,
c
if c = 0.

If also B Gl(2, C), a calculation gives

(22.9)

LA LB = LAB .

In particular LA is bijective in (22.6), with inverse LA1 . If we give C {} its natural topology as the one-point compactification of C, we have LA a homeomorphism in
(22.6). Later we will give C {} the structure of a Riemann surface and see that LA is
biholomorphic on this surface.
Note that LsA = LA for any nonzero s C. In particular, LA = LA1 for some A1
of determinant 1; we say A1 Sl(2, C). Given Aj Sl(2, C), LA1 = LA2 if and only
if A2 = A1 . In other words, the group of linear fractional transformations (22.5) is
isomorphic to
(22.10)

P Sl(2, C) = Sl(2, C)/(I).

Note that if a, b, c, d are all real, then LA in (22.5) preserves R {}. In this case we
have A Gl(2, R). We still have LsA = LA for all nonzero s, but we need s R to get
sA Gl(2, R). We can write LA = LA1 for A1 Sl(2, R) if A Gl(2, R) and det A > 0.
We can also verify that
(22.11)

A Sl(2, R) = LA : U U,

where U = {z : Im z > 0} is the upper half-plane. In fact, for a, b, c, d R, z = x + iy,

az + b
(az + b)(cz + d)
=
= R + iy
,
cz + d
(cz + d)(cz + d)
P
with R R, P = |cz + d|2 > 0, which gives (22.11). Again LA = LA , so the group
(22.12)
acts on U.

P Sl(2, R) = Sl(2, R)/(I)

164

We now single out for attention the following linear fractional transformation:

zi
1 i
(22.13)
(z) =
, (z) = LA0 (z), A0 =
.
1 i
z+i
It is clear that
(22.14)

: R {} S 1 = D,

: U D,

where D = {z : |z| < 1} is the unit disk. Note that conjugating the Sl(2, R) action on U
by yields the mappings
MA = LA0 AA1 : D D.

(22.15)

In detail, if A is as in (5), with a, b, c, d real, and if A0 is as in (22.13),

1 (a + d)i b + c (a d)i + b + c
1
(22.16)
A0 AA0 =
.
2i (a d)i b c (a + d)i + b c
It follows that
(22.17)

A0 Sl(2, R) A1
0

n
=

o
Gl(2, C) : ||2 ||2 = 1 .

Hence we have linear fractional transformations

z +

,
(22.18)
LB : D D, B =
, LB (z) =

z +

||2 ||2 = 1.

The group described on the right side of (22.17) is denoted SU (1, 1). Note that for such
B as in (22.18),
ei +
+ ei
LB (ei ) = i
= ei
,
e +
+ ei
and in the last fraction the numerator is the complex conjugate of the denominator. This
directly implies the result LB : D D for such B.
We have the following important transitivity properties.
Proposition 22.2. The group SU (1, 1) acts transitively on D, via (22.18). Hence Sl(2, R)
acts transitively on U via (22.5).
Proof. Given p D, we can pick , C such that ||2 ||2 = 1 and / = p. Then
LB (0) = p. This together with LB1 L1
B2 = LB1 B21 is enough to establish transitivity of
SU (1, 1) on D. As for transitivity of Sl(2, R) on U, one can use the conjugation (22.17),
or the following direct argument. Given p = a + ib U (a R, b > 0), Lp (z) = bz + a =
(b1/2 z + b1/2 a)/b1/2 maps i to p.
The following converse to Proposition 22.2 is useful.

165

Proposition 22.3. If f : D D is a holomorphic diffeomorphism, then f = LB for

some B SU (1, 1). Hence if F : U U is a holomorphic diffeomorphism, then F = LA
for some A Sl(2, R).
Proof. Say f (0) = p D. By Proposition 22.2 there exists B1 SU (1, 1) such that
LB1 (p) = 0, so g = LB1 f : D D is a holomorphic diffeomorphism satisfying g(0) = 0.
Now we claim that g(z) = cz for a constant c with |c| = 1.
To see this, note that, h(z) = g(z)/z has a removable singularity at 0 and yields a
holomorphic map h : D C. A similar argument applies to z/g(z). Furthermore, with
= {z D : |z| = } one has, because g : D D is a homeomorphism,
lim sup |g(z)| = lim inf |g(z)| = 1.
%1 z

%1 z

Hence the same can be said for h| , and then a maximum principle argument yields
|g(z)/z| 1 on D and also |z/g(z)| 1 on D; hence |g(z)/z| 1 on D. This implies
g(z)/z = c, a constant, and that |c| = 1, as asserted.

c 0
To proceed, we have g = LB2 with B2 = a
, and we can take a = c1/2 , to
0 1
obtain B2 SU (1, 1). We have f = LB 1 B2 , and the proof is complete.
1

We single out some building blocks for the group of linear fractional transformations,
namely (with a 6= 0)
1
.
z
We call these respectively (complex) dilations, translations, and inversion about the unit
circle {z : |z| = 1}. These have the form (22.5), with A given respectively by

a 0
1 b
0 1
(22.20)
,
,
.
0 1
0 1
1 0
(22.19)

a (z) = az,

b (z) = z + b,

(z) =

We can produce the inversion D about the boundary of a disk D = Dr (p) as

(22.21)

D = p r 1/r p .

The reader can work out the explicit form of this linear fractional transformation. Note
that D leaves D invariant and interchanges p and .
Recall that the linear fractional transformation in (22.13) was seen to map R {}
to S 1 . Similarly its inverse, given by i(z) with
(22.22)

(z) =

z+1
,
z1

maps S 1 to R {}; equivalently maps S 1 to iR {}. To see this directly, write

(22.23)

(ei ) =

ei + 1
i
=
.
i
e 1
tan /2

These are special cases of an important general property of linear fractional transformations. To state it, let us say that an extended line is a set ` {}, where ` is a line in
C.

166

Proposition 22.4. If L is a linear fractional transformation, then L maps each circle to

a circle or an extended line, and L maps each extended line to a circle or an extended line.
To begin the proof, suppose D C is a disk. We investigate where L maps D.
Claim 1. If L has a pole at p D, then L maps D to an extended line.
Proof. Making use of the transformations (22.19), we have L(D) = L0 (S 1 ) for some linear
fractional transformation L0 , so we need check only the case D = {z : |z| < 1}, with L
having a pole on S 1 , and indeed we can take the pole to be at z = 1. Thus we look at
aei + b
ei 1
a+b
i
ab
=
+
,
2 tan /2
2

L(ei ) =
(22.24)

whose image is clearly an extended line.

Claim 2. If L has no pole on D, then L maps D to a circle.
Proof. One possibility is that L has no pole in C. Then c = 0 in (22.5). This case is
elementary.
Next, suppose L has a pole at p D. Composing (on the right) with various linear
fractional transformations, we can reduce to the case D = {z : |z| < 1}, and making
further compositions (via Proposition 22.2), we need only deal with the case p = 0. So we
are looking at
(22.25)

L(z) =

az + b
,
z

L(ei ) = a + bei .

Clearly the image L(S 1 ) is a circle.

If L has a pole at p C \ D, we can use an inversion about D to reduce the study to
that done in the previous paragraph. This finishes Claim 2.
To finish the proof of Proposition 22.4, there are two more claims to establish:
Claim 3. If ` C is a line and L has a pole on `, or if L has no pole in C, then L maps
` {} to an extended line.
Claim 4. If ` C is a line and L has a pole in C \ `, then L maps ` {} to a circle.
We leave Claims 34 as exercises for the reader.
We present a variety of examples of conformal maps in Figs. 22.122.3. The domains
pictured there are all simply connected domains, and one can see that they are all conformally equivalent to the unit disk. The Riemann mapping theorem, which we will prove
in the next section, says that any simply connected domain C such that 6= C is
conformally equivalent to the disk. Here we make note of the following.

167

Proposition 22.5. If C is simply connected and 6= C, then there is a holomorphic

diffeomorphism f : O, where O C is a bounded, simply connected domain.
Proof. Pick p C \ and define a holomorphic branch on of
g(z) = (z p)1/2 .

(22.26)

The simple connectivity of guarantees the existence of such g, as shown in 20. Now
e
g is one-to-one on and it maps diffeomorphically onto a simply connected region
having the property that
e = z
e
(22.27)
z
/ .
e and if we compose g with inversion across D
It follows that there is a disk D C \ ,
we obtain such a desired holomorphic diffeomorphism.

Exercises
1. Find conformal mappings of each of the following regions onto the unit disk. In each
case, you can express the map as a composition of various conformal maps.
(a) = {z = x + iy : y > 0, |z| > 1}.

(b) = C \ (, 1] [1, ) .
(c) = {z : |z + 1/2| < 1} {z : |z 1/2| < 1}.
2. Consider the quarter-plane
= {z = x + iy : x > 0, y > 0}.
Find a conformal map of onto itself such that
(1 + i) = 2 + i.
3. Let f : O be a conformal diffeomorphism. Show that if u : O R is harmonic, so
is u f : R.
4. Write out the details to establish Claims 34 in the proof of Proposition 22.4.
5. Reconsider Exercise 1b) of 21, mapping the region U defined there conformally onto
the unit disk.
6. Given q D = {z : |z| < 1}, define
zq
.
1 qz
Show that q : D D and q (q) = 0. Write q in the form (22.18). Relate this to the
proof of Proposition 22.2.

(22.28)

q (z) =

168

23. The Riemann mapping theorem

We make the standing convention that a domain C is nonempty, open, and connected. Our aim in this section is to establish the following.
Theorem 23.1. Assume C is a simply connected domain, and 6= C. Then there
exists a holomorphic diffeomorphism
(23.1)

f : D

of onto the unit disk D = {z C : |z| < 1}.

The proof given here is due to P. Koebe. To begin the proof, we recall from 22 that
we know is conformally equivalent to a bounded domain in C, so it suffices to treat the
bounded case. Thus from here on we assume is bounded. Fix p .
We define F to be the set of holomorphic maps g : D that have the following three
properties:
(i) g is one-to-one (we say univalent),
(ii) g(p) = 0,
(iii) g 0 (p) > 0.
For C bounded it is clear that b(z p) belongs to F for small b > 0, so F is nonempty.
Note that if R = dist(p, ), then, by (5.32),
(23.2)

|g 0 (p)|

1
,
R

g F,

so we can set
(23.3)

and we have A < . Pick g F such that g0 (p) A as . A normal family

argument from 21 shows that there exists a subsequence g f locally uniformly on ,
and
(23.4)

f : D

is holomorphic and satisfies f 0 (p) = A. We claim this function provides the holomorphic
diffeomorphism (23.1). There are two parts to showing this, treated in the next two
lemmas.
Lemma 23.2. In (23.4), f is one-to-one.
Proof. Suppose there exist distinct z1 , z2 such that f (z1 ) = f (z2 ) = w D. Let
O be a smoothly bounded region such that z1 , z2 O and such that f () is disjoint
from w, where = O.
By the argument principle (Proposition 17.4), f () winds (at least) twice about w. But
each g () winds only once about w. Since g f uniformly on , this is a contradiction.

169

Lemma 23.3. In (23.4), f is onto.

Proof. Suppose f () omits q D. Form the holomorphic function
s
f (z) q
(23.5)
F (z) =
.
1 qf (z)
It is here that we use the hypothesis that is simply connected, to guarantee the existence
of such a square root. We have
(23.6)

F : D.

(23.7)

G(z) =

|F 0 (p)| F (z) F (p)

.
F 0 (p) 1 F (p)F (z)

(23.8)

G : D.

Also, G(p) = 0, and a computation gives

(23.9)

G0 (p) =

|F 0 (p)|
1 + |q|
= p A > A.
2
1 |F (p)|
2 |q|

Hence G F and G0 (p) > A, a contradiction. This proves Lemma 23.3, and hence the
Riemann mapping theorem.

Exercises
1. Note that F (z) in (23.5) is given by
q
(23.10)

F (z) =

q f (z) ,

(23.11)

G(z) =

|F 0 (p)|
F (p) F (z) .
0
F (p)

Use these identities and the chain rule to verify (23.9).

2. Suppose h : D D are holomorphic, univalent maps satisfying
(23.12)

h (0) = 0,

h0 (0) > 0,

h (D) D ,

1.

170

(23.13)

Then show that

(23.14)

h (z) z

locally uniformly on D.

Hint. Use a normal families argument, and show that any limit hk g must have the
property that g(D) = D, and conclude that g(z) = z. (The argument principle may be
useful.)
3. Suppose is a bounded, simply connected domain, p , and f : D are univalent
holomorphic maps satisfying
(23.15)

f (p) = 0,

f0 (p) > 0,

f () D ,

1.

Show that f f locally uniformly on , where f : D is the Riemann mapping

function given by Theorem 23.1.
Hint. Consider h = f f 1 : D D.
4. Let f1 : D be a univalent holomorphic mapping satisfying f1 (p) = 0, f10 (p) = A1 >
0 (i.e., an element of F). Asuming f1 is not onto, choose q1 D \ f1 () with minimal
possible absolute value. Construct f2 F as
(23.16)

f2 (z) =

|F10 (p)|
F1 (p) F1 (z) ,
0
F1 (p)

F1 (z) =

q1 f1 (z) .

Evaluate A2 = f20 (p). Take q2 D \ f2 () with minimal absolute value and use this to
construct f3 . Continue, obtaining f4 , f5 , . . . . Show that at least one of the following holds:
(23.17)

f0 (p) A,

or f () D ,

1,

with A as in (23.3). Deduce that f f locally uniformly on , where f : D is the

Riemann mapping function.
Hint. If |q1 | is not very close to 1, then A2 is somewhat larger than A1 . Similarly for A+1
compared with A .

171

24. Boundary behavior of conformal maps

Throughout this section we assume that C is a simply connected domain and
f : D is a holomorphic diffeomorphism, where D = {z : |z| < 1} is the unit disk. We
look at some cases where we can say what happens to f (z) as z approaches the boundary
. The following is a simple but useful result.
Lemma 24.1. We have
(24.1)

z = |f (z)| 1.

Proof. For each > 0, D1 = {z : |z| 1 } is a compact subset of D, and K =

f 1 (D1 ) is a compact subset of . As soon as z
/ K , |f (z)| > 1 .
We now obtain a local regularity result.
Proposition 24.2. Assume : (a, b) C is a simple real analytic curve, satisfying
0 (t) 6= 0 for all t. Assume is part of , with all points near to and on the left side of
(with its given orientation) belonging to . Then there is a neighborhood V of in C and
a holomorphic extension F of f to F : V C. We have F () D and F 0 () 6= 0
for all .
Proof. There exists a neighborhood O of (a, b) in C and a univalent holomorphic map
: O C extending . Say V = (O). See Fig. 24.1. We can assume O is symmetric
with respect to reflection across R. Say O = { O : Im > 0}.
We have f : O+ D and
(24.2)

z O+ , z L = O R = |f (z )| 1.

It follows from the form of the Schwarz reflection principle given in 13 that g = f |O+
has a holomorphic extension G : O C, and G : L D. Say U = G(O), as in Fig. 24.1.
Note that U is invariant under z 7 z 1 .
Then we have a holomorphic map
(24.3)

F = G 1 : V U.

It is clear that F = f on V . It remains to show that F 0 () 6= 0 for . It is

equivalent to show that G0 (t) 6= 0 for t L. To see this, note that G is univalent on O+ ;
G|O+ = g|O+ : O+ D. Hence G is univalent on O ; G|O : O C \ D. The argument
principle then gives G0 (t) 6= 0 for t L. This finishes the proof.
Using Proposition 24.2 we can show that if is real analytic then f extends to a
homeomorphism from to D. We want to look at a class of domains with non-smooth
boundaries for which such a result holds. Clearly a necessary condition is that be
homeomorphic to S 1 , i.e., that be a Jordan curve. C. Caratheodory proved that this
condition is also sufficient. A proof can be found in [Ts]. Here we establish a simpler
result, which nevertheless will be seen to have interesting consequences.

172

Proposition 24.3. In addition to the standing assumptions on , assume it is bounded

and that is a simple closed curve that is a finite union of real analytic curves. Then
the Riemann mapping function f extends to a homeomorphism f : D.
Proof. By Proposition 24.2, f extends to the real analytic part of , and the extended
f maps these curves diffeomorphically onto open intervals in D. Let J1 and J2 be real
analytic curves in , meeting at p, as illustrated in Fig. 24.2, and denote by I the images
in D. We claim that I1 and I2 meet, i.e., the endpoints q1 and q2 pictured in Fig. 24.2
coincide.
Let r be the intersection {z : |z p| = r}, and let `(r) be the length of f (r ) = r .
Clearly |q1 q2 | `(r) for all (small) r > 0, so we would like to show that `(r) is small
for (some) small r. R
We have `(r) = r |f 0 (z)| ds, and Cauchys inequality implies
`(r)2
2
r

(24.4)

Z
|f 0 (z)|2 ds.
r

If `(r) for r R, then integrating over r [, R] yields

ZZ

R
2
|f 0 (z)|2 dx dy = 2 Area f (, R) 2 2 ,
(24.5)
log 2

(,R)

where (, R) = {z : |z p| R}. Since log(1/) as & 0, there exists

arbitrarily small r > 0 such that `(r) < . Hence |q1 q2 | < , so q1 = q2 , as asserted.
It readily follows that taking f (p) = q1 = q2 extends f continuously at p. Such an
extension holds at other points of where two real analytic curves meet, so we have a
continuous extension f : D. This map is also seen to be one-to-one and onto. Since
and D are compact, this implies it is a homeomorphism, i.e., f 1 : D is continuous
(cf. Exercise 9 below).

Exercises
1. Suppose f : D is a holomorphic diffeomorphism, extending to a homeomorphism
f : D. Let g C() be given. Show that the Dirichlet problem

(24.6)
u = 0 in , u = g
has a unique solution u C 2 () C(), given by u = v f , where

(24.7)
v = 0 in D, v D = g f 1 D ,
the solution to (24.7) having been given in 13.

173

2. Verify that for f : C holomorphic, if we consider Df (z) End(R2 ), then

det Df (z) = |f 0 (z)|2 , and explain how this yields the identity (24.5).
3. Let C satisfy the hypotheses of Proposition 24.3. Pick distinct p1 , p2 , p3
such that, with its natural orientation, runs from p1 to p2 to p3 , and back to p1 . Pick
q1 , q2 , q3 D with the analogous properties. Show that there exists a unique holomorphic
diffeomorphism f : D whose continuous extension to takes pj to qj , 1 j 3.
Hint. First tackle the case = D.
In Exercises 46, pick p > 0 and let R C be the rectangle with vertices at 1, 1, 1 + ip,
and 1 + ip. Let : R D be the Riemann mapping function such that
(24.8)

(1) = i,

(0) = 1,

(1) = i.

Define : R U (the upper half plane) by

(24.9)

(z) = i

(z) 1
.
(z) + 1

4. Show that (ip) = 1 (so (z) as z ip). Show that extends continuously to
R \ {ip} C and
(24.10)

(1) = 1,

(0) = 0,

(1) = 1.

5. Show that you can apply the Schwarz reflection principle repeatedly and extend to a
meromorphic function on C, with simple poles at ip + 4k + 2i`p, k, ` Z. Show that
(24.11)

(z + 4) = (z + 2ip) = (z).

Hint. To treat reflection across the top boundary of R, apply Schwarz reflection to 1/.
Remark. We say is doubly periodic, with periods 4 and 2ip.
6. Say (1 + ip) = r. Show that r > 0, that (1 + ip) = r, and that
(24.12)

ip
2

z =

r
.
(z)

7. Let T C be the equilateral triangle with vertices at 1, 1, and

be the holomorphic diffeomorphism with boundary values
(24.13)

(1) = 1,

(0) = 0,

3i. Let : T U

(1) = 1.

Use Schwarz reflection to produce a meromorphic extension of to C, which is doubly

periodic. Show that

174

What are the poles of ? Cf. Fig. 24.3.

8. In the context of Exercise 7, show that

(i 3) = .
Let

# + z = + e2i/3 z .
3
3

# (1) = 1,
Conclude that

# (1) = ,

# (i 3) = 1.

# (z) = (z),

where : U U is the holomorphic diffeomorphism satisfying

(1) = 1,

(1) = ,

so
(z) =

() = 1,

z+3
.
z1

9. Let X and Y be compact metric spaces, and assume f : X Y is continuous, one-toone, and onto. Show that f is a homeomorphism, i.e., f 1 : Y X is continuous.
Hint. You are to show that if f (xj ) = yj y, then xj f 1 (y). Now since X is compact,
every subsequence of (xj ) has a further subsequence that converges to some point in X...

175

25. The disk covers C \ {0, 1}

Our main goal in this section is to prove that the unit disk D covers the complex plane
with two points removed, holomorphically. Formally:
Proposition 25.1. There exists a holomorphic covering map
(25.1)

: D C \ {0, 1}.

The proof starts with an examination of the following domain . It is the subdomain
of the unit disk D whose boundary consists of three circular arcs, intersecting D at right
angles, at the points {1, e2i/3 , e2i/3 }. See Fig. 25.1. If we denote by
(25.2)

: D U = {z C : Im z > 0}

the linear fractional transformation of D onto U with the property that (1) = 0, (e2i/3 )
e = () is pictured in Fig. 25.2.
= 1, and (e2i/3 ) = , the image
The Riemann mapping theorem guarantees that there is a holomorphic diffeomorphism
(25.3)

: D,

and by Proposition 24.3 this extends to a homeomorphism : D. We can take to

leave the points 1 and e2i/3 fixed. Conjugation with the linear fractional transformation
gives a holomorphic diffeomorphism
(25.4)

e U,
= 1 :

e onto the real axis, with (0) = 0 and (1) = 1.

and extends to map
Now the Schwarz reflection principle can be applied to , reflecting across the vertical
e to extend to the regions O
e2 and O
e3 in Fig. 25.2. A variant extends to O
e1 .
lines in ,
e O
e1 O
e2 O
e3
(Cf. Exercise 1 in 8.) Note that this extension maps the closure in U of
onto C \ {0, 1}. Now we can iterate this reflection process indefinitely, obtaining
(25.5)

: U C \ {0, 1}.

Furthermore, this is a holomorphic covering map. Then = gives the desired

holomorphic covering map (25.1).

176

Exercises
1. Show that the map : D U in (25.2) is given by
(z) =

z1
,
z 2

= e2i/3 .

Show that
() = 1,

( 2 ) =

1
.
2

For use below, in addition to z 7 z, we consider the following anti-holomorphic involutions

of C: z 7 z , z 7 z , z 7 z , and z 7 z c , given by

1 z
+z = + ,
2
2
4
1
c
1
(x + iy) = x + iy,
+ z = + z.
2
2
z =

(25.6)

1
,
z

(25.7)

(z ) = (z),

(z ) = (z),

(z + 2) = (z).

3. Show that
(25.8)

(z c ) = (z)c .

e invariant, and that

4. Show that z 7 z leaves
(25.9)

(z ) = (z) .

e Use Exercise 1 to show that (25.9) is equivalent

Hint. First establish this identity for z .
to the statement that in (25.3) commutes with reflection across the line through 0 and
e2i/3 , while (25.8) is equivalent to the statement that commutes with reflection across
the line through 0 and e2i/3 .
5. Show that z = 1/z and (z ) = 1/z. Deduce from (25.7) and (25.9) that
(25.10)

1
1

=
.
z
(z)

177

6. Show that (z )c = z + 1 and (z)c = 1 z. Deduce from (25.7) and (25.8) that
(25.11)

(z + 1) = 1 (z).

As preparation for Exercises 79, the reader should peek at 26.

7. Show that F01 , F0 : S 2 S 2 (S 2 = C {}), given by
(25.12)

F01 (w) = 1 w,

F0 (w) =

1
,
w

are holomorphic automorphisms of S 2 that leave C \ {0, 1} invariant, F01 fixes and
switches 0 and 1, while F0 fixes 1 and switches 0 and . Show that these maps generate
a group G of order 6, of automorphisms of S 2 and of C \ {0, 1}, that permutes {0, 1, }
and is isomorphic to the permutation group S3 on three objects. Show that
(25.13)

F1 (w) = F0 F01 F0 (w) =

w
w1

is the element of G that fixes 0 and switches 1 and . Show that the rest of the elements
of G consist of the identity map, w 7 w, and the following two maps:
1
,
1w
w1
.
F10 (w) = F01 F0 (w) =
w
F01 (w) = F0 F01 (w) =
(25.14)

8. Show that the transformations in G listed in (25.12)(25.14) have the following fixed
points.
Element
F0
F01
F1
F01
F10

Fixed points
1, 1 = A1
, 12 = A2
0, 2 = A3
ei/3 = B
ei/3 = B

See Figure 25.3. Show that the elements of G permute {A1 , A2 , A3 } and also permute
{B+ , B }.
9. We claim there is a holomorphic map
(25.15)

H : S 2 S 2 ,

178

satisfying
(25.16)

F G,

such that
(25.17)

H(ei/3 ) = 0,

(25.18)

with zeros of order 3 at each of these points, and

H(1) = H( 12 ) = H(2) = 1,

(25.19)

H(w) 1 having zeros of order 2 at each of these points.

To obtain the properties (25.17)(25.18), we try
(25.20)

H(w) = C

(w ei/3 )3 (w ei/3 )3
(w2 w + 1)3
=
C
,
w2 (w 1)2
w2 (w 1)2

(25.21)

C=

4
,
27

so
(25.22)

H(w) =

4 (w2 w + 1)3
.
27 w2 (w 1)2

Verify that
(25.23)

1
w

= H(w),

and show that (25.16) follows.

Remark. The map in (25.5) is a variant of the elliptic modular function, and the
composition H is a variant of the j-invariant. Note from (25.10)(25.11) that
(25.24)

1
H
= H (z + 1) = H (z),
z

z U.

For more on these maps, see the exercises at the end of 34. For a different approach, see
the last two sections in Chapter 7 of [Ahl].

179

26. The Riemann sphere and other Riemann surfaces

Our main goal here is to describe how the unit sphere S 2 R3 has a role as a conformal
compactification of the complex plane C. To begin, we consider a map
S : S 2 \ {e3 } R2 ,

(26.1)

known as stereographic projection; here e3 = (0, 0, 1). We define S as follows:

(26.2)

S(x1 , x2 , x3 ) = (1 x3 )1 (x1 , x2 ).

(26.3)

S 1 (x, y) =

1
(2x, 2y, r2 1),
1 + r2

r 2 = x2 + y 2 .

The following is a key geometrical property.

Proposition 26.1. The map S is a conformal diffeomorphism of S 2 \ {e3 } onto R2 .
In other words, we claim that if two curves 1 and 2 in S 2 meet at an angle at
p 6= e3 , then their images under S meet at the same angle at q = S(p). It is equivalent,
and slightly more convenient, to show that F = S 1 is conformal. We have
(26.4)

DF (q) : R2 Tp S 2 R3 .

See Appendix C for more on this. Conformality is equivalent to the statement that there
is a positive function (p) such that, for all v, w R2 ,
(26.5)

DF (q)v DF (q)w = (q) v w,

or in other words,

(26.6)

1 0
0 1

To check (26.6), we compute DF via (26.3). A calculation gives

1 x2 + y 2
2xy
2
2xy
(26.7)
DF (x, y) =
1 + x2 y 2 ,
2
2
(1 + r )
2x
2y
and hence
(26.8)

4
DF (x, y) DF (x, y) =
(1 + r2 )2
t

1
0

0
1

180

This gives Proposition 26.1.

Similarly we can define a conformal diffeomorphism
(26.9)

S : S 2 \ {e3 } R2 .

To do this we take x3 7 x3 . This reverses orientation, so we also take x2 7 x2 . Thus

we set
(26.10)

S (x1 , x2 , x3 ) = (1 + x3 )1 (x1 , x2 ).

(26.11)

S S 1 (x, y) =

1
(x, y).
r2

(26.12)

S S 1 (z) =

1
z
= .
2
|z|
z

Clearly the composition of conformal transformations is conformal, so we could predict in

advance that S1 S 1 would be conformal and orientation-preserving, hence holomorphic,
and (26.12) bears this out.
If we use (26.1) to identify C with S 2 \{e3 }, then the one-point compactification C{}
is naturally identified with S 2 , with corresponding to the north pole e3 . The map
(26.12) can be extended from C \ {0} to C {}, and it switches 0 and .
The concept of a normal family of maps S, introduced in 21, is of great interest
when S = S 2 = C {}. The following result produces a key link with results established
in 21.
Proposition 26.2. Assume C is a connected open set. A family F of holomorphic
functions C is normal with respect to (, C {}) if and only if for each sequence
f from F one of the following happens:
(a) A subsequence fk converges uniformly on each compact K , as a sequence fk :
K C, or
(b) A subsequence fk tends to uniformly on each compact K .
Proof. Assume F is a normal family with respect to (, C {}), and f is a sequence
of elements of F. Take a subsequence fk , uniformly convegent on each compact K, as a
sequence of maps fk : K S 2 . Say fk f : S 2 . Pick p . We consider two
cases.
Case I. First suppose f (p) = . Then there exists N Z+ and a neighborhood U of p
in such that |fk (z)| 1 for z U, k N . Set gk (z) = 1/fk (z), for z U, k N .
We have |gk | 1 on U , gk (z) 6= 0, and gk (z) 1/f (z), locally uniformly on U (with
1/ = 0), and in particular gk (p) 0. By Hurwitz theorem (Proposition 17.8), this

181

implies 1/f (z) = 0 on all of U , i.e., f = on U , hence f = on . Hence Case I

Case (b).
Case II. Suppose f (p) C, i.e., f (p) S 2 \ {}. By the analysis in Case I it follows that
f (z) C for all z . It is now straightforward to verify Case (a) here.
This gives one implication in Proposition 26.2. The reverse implication is easily established.
The surface S 2 is an example of a Riemann surface, which we define as follows. A
Riemann surface is a two-dimensional manifold M covered by open sets Oj with coordinate
charts j : j Oj having the property that, if Oj Ok 6= , and if jk = 1
j (Oj Ok ),
then the diffeomorphism
1
k j : jk kj

(26.13)

is holomorphic. See Appendix C for general background on manifolds and coordinate

charts.
Another important class of Riemann surfaces is given as follows. Let R2 C be
the image of Z2 under any matrix in Gl(2, R). Then the torus
(26.14)

T = C/

is a Riemann surface in a natural fashion.

There are many other Riemann surfaces. For example, any oriented two-dimensional
Riemannian manifold has a natural structure of a Riemann surface. A proof of this can
be found in Chapter 5 of [T2]. An important family of Riemann surfaces holomorphically
diffeomorphic to surfaces of the form (26.14) will arise in 34, with implications for the
theory of elliptic functions.

Exercises
1. Give an example of a family F of holomorphic functions C with the following two
properties:
(a) F is normal with respect to (, S 2 ).
(b) {f 0 : f F} is not normal, with respect to (, C).
Compare Exercise 2 of 21. See also Exercise 11 below.
2. Given C open, let
F = {f : C : Re f > 0 on , f holomorphic}.
Show that F is normal with respect to (, S 2 ). Is F normal with respect to (, C)?

182

3. Let F = {z n : n Z+ }. For which regions is F normal with respect to (, S 2 )?

Compare Exercise 4 in 21.
4. Show that the set of orientation-preserving conformal diffeomorphisms : S 2 S 2 is
precisely the set of linear fractional transformations of the form (22.5), with A Gl(2, C).
Hint. Given such : S 2 S 2 , take LA such that LA takes to , so = LA |S 2 \{}
is a holomorphic diffeomorphism of C onto itself. What form must have? (Cf. Proposition
11.4.)
5. There is a natural notion of when a map : M1 M2 between two Riemann surfaces
is holomorphic. Write it down. Show that if and also : M2 M3 are holomorphic,
then so is : M1 M3 .
6. Let p(z) and q(z) be polynomials on C. Assume the roots of p(z) are disjoint from the
roots of q(z). Form the meromorphic function
R(z) =

p(z)
.
q(z)

Show that R(z) has a unique continuous extension R : S 2 S 2 , and this is holomorphic.
Exercises 79 deal with holomorphic maps F : S 2 S 2 . Assume F is not constant.
7. Show that there are only finitely many pj S 2 such that DF (pj ) : Tpj S 2 Tqj S 2 is
singular (hence zero), where qj = F (pj ). The points qj are called critical values of F .
8. Suppose is not a critical value of F and that F 1 () = {, p1 , . . . , pk }. Show that
f (z) = F (z)(z p1 ) (z pk ) : C C,
and |f (z)| as |z| . Deduce that f (z) is a polynomial in z. (Cf. Proposition
11.4.)
9. Show that every holomorphic map F : S 2 S 2 is of the form treated in Exercise 6
(except for the constant map F ).
Hint. Compose with linear fractional transformations and transform F to a map satisfying
the conditions of Exercise 8.
10. Given a holomorphic map f : C, set
(26.13)

g = S 1 f : S 2 .

For z , set q = f (z), p = g(z), and consider

(26.14)

Dg(z) : R2 Tp S 2 .

183

(26.15)

Dg(z)t Dg(z) = 4

|f 0 (z)| 2
I,
1 + |f (z)|2

f # (z) =

(26.16)

|f 0 (z)|
1 + |f (z)|2

is sometimes called the spherical derivative of f .

11. Using (26.15), show that a family F of holomorphic functions on is normal with
respect to (, S 2 ) if and only if for each compact K ,
(26.17)

{f # (z) : f F, z K} is bounded.

Hint. Check Proposition 21.1.

12. Show that the meromorphic function constructed in Exercises 46 of 24 yields a
holomorphic map
: T S 2 ,

(26.18)
where = {4k + 2i`p : k, ` Z}.

Changing notation in (26.1), let us write

(26.19)

S : S 2 C {},

bijective,

with S(e3 ) = .
13. For R, define : C C by (z) = ei z. Also set () = . Show that

cos sin
.
(26.20)
R = S 1 S = R = sin
cos
1
14. Let R : S 2 S 2 be a conformal diffeomorphism with the properties
R(e3 ) = e3 ,

R : E E,

where E = {(x1 , x2 , x3 ) S 2 : x3 = 0}. Show that R = R (as in 26.20)) for some R.

Hint. Consider = S R S 1 . Show that : C C is bijective and preserves
{z C : |z| = 1}. Deduce that = for some R.

184

(26.21)

f (z) =

(cos )z sin
,
(sin )z + cos

f : C {} C {}.

Set
= S 1 f S,

(26.22)

: S 2 S 2 .

Show that is a conformal diffeomorphism.

16. In the setting of Exercise 15, show that, for all R,
(e2 ) = e2 ,

e E,
e
: E

e = {(x1 , x2 , x3 ) S 2 : x2 = 0}. Show also that

where E
(e3 ) = (sin 2, 0, cos 2).
Hint. To get started, show that
f (i) = i,

f : R {} R {},

f () =

cos
.
sin

17. In the setting of Exercises 1516, show that

cos 2
sin 2
.
(26.23)
=
1
sin 2
cos 2
Hint. Translate the result of Exercise 14 to this setting.
To A G`(2, C) we associate the linear fractional transformation LA as in (22.5),
(26.24)

az + b
LA (z) =
,
cz + d

A=

a b
c d

LA : C {} C {}.

As also define
(26.25)

A = S 1 LA S,

A : S 2 S 2 .

18. Show that if A G`(2, C), A : S 2 S 2 is a holomorphic diffeomorphism. If also

B G`(2, C), AB = A B .

185

Exercise 4 says the class of holomorphic diffeomorphisms : S 2 S 2 is equal to the class

of maps A as A runs over G`(2, C).
19. Given 3 distinct points a, b, c C, show that there exists a linear fractional transformation
za
L(z) =
such that L(a) = 0, L(b) = 1, L(c) = .
zc
Deduce that if {p, q, r} and {p0 , q 0 , r0 } are two sets of 3 distinct points in S 2 , then there
exists a holomorphic diffeomorphism : S 2 S 2 such that
(p) = p0 ,

(q) = q 0 ,

(r) = r0 .

Show that such is unique.

20. Let be a circle in S 2 . Show that
S is a circle in C if e3
/ , and
S is an exended line in C {} if e3 .
Hint. For the first part, take a rotation R such that 0 = R is a circle centered at e3 . Show
1
directly that S0 = is a circle in C. Deduce that S = L1
is
A where LA = S R S
a linear fractional transformation. Then apply Proposition 22.4 to L1
.
A
21. Let C be a circle. Show that S 1 () is a circle in S 2 .
Hint. Pick a and let p = S 1 (a) S 2 . Pick a rotation R such that R(p) = e3 , so
R S 1 (a) = e3 . Now = R S 1 () is a curve in S 2 , and we want to show that it is a
circle.
Indeed, S() = S R S 1 () = L(), with L a linear fractional transformation. L()
contains S(e3 ) = , so S() = L() = ` is an extended line (by Proposition 22.4). Then
= S 1 (`), which is seen to be a circle in S 2 . In fact, S 1 (`) is the intersection of S 2 and
the plane through ` and e3 .
22. Show that if is a circle in S 2 and : S 2 S 2 is a holomorphic diffeomorphism, then
() is a circle in S 2 .
Hint. Use Exercises 2021 and Proposition 22.4.

186

27. Montels theorem

Here we establish a theorem of Montel, giving a highly nontrivial and very useful sufficient condition for a family of maps from a domain to the Riemann sphere S 2 to be a
normal family.
Theorem 27.1. Fix a connected domain U C and let F be the family of all holomorphic
maps f : U S 2 = C {} with range in S 2 \ {0, 1, }. Then F is a normal family.
Proof. There is no loss of generality in treating the case U = D, the unit disk; in particular
we can assume henceforth that U is simply connected.
Take f F, and let K be any compact connected subset of U . We aim to show that
some subsequence fk converges uniformly on K. Two cases arise:
Case A. There exist pk K such that fk (pk ) is bounded away from {0, 1, } in S 2 .
Case B. Such a subsequence does not exist.
We can dispose of Case B immediately. Using the connectedness of K, we must have a
subsequence fk converging to either 0, 1, or uniformly on K.
It remains to deal with Case A. First, to simplify the notation, relabel k as . We
make use of the holomorphic covering map : D C \ {0, 1} given in 25, and refer to
Fig. 25.1 in that section. Pick
( O1 O2 O3 ) D
such that ( ) = f (p ). At this point it is crucial to observe that | | 1 for some
> 0, independent of . Now we take the unique lifting g : U D of f such that
g (p ) = . That it is a lifting means f = g . The existence of such a lifting follows
from the hypothesized simple connectivity of U .
The uniform boundedness of {g } implies that a subsequence (which again we relabel
g ) converges locally uniformly on U ; we have g g : U D. Furthermore, again
passing to a subsequence, we can assume p p K and
(27.1)

g (p ) = D1 .

(27.2)

g g : U D.

(27.3)

g (K) D1 ,

This in turn gives the desired convergence f g, uniformly on K.

The last argument shows that in Case A the limit function f = g maps U to
S \ {0, 1, }, so we have the following. (Compare Proposition 26.2.)
2

187

Corollary 27.2. In the setting of Theorem 27.1, if f F and f f locally uniformly,

then either
(27.4)

f 0,

f 1,

or f : U S 2 \ {0, 1, }.

f ,

Exercises on Fatou sets and Julia sets

Let R : S 2 S 2 be holomorphic, having the form R(z) = p(z)/q(z) with p(z) and q(z)
polynomials with no common zeros. We set d = deg R = max {deg p, deg q}, called the
degree of the map R.
1. Show that if p1 S 2 is not a critical value of R, then R1 (p1 ) consists of d points.
2. Define R2 = R R, R3 = R R2 , . . . , Rn = R Rn1 . Show that deg Rn = dn .
3. Show that if d 2 then {Rn : n 1} is not a normal family of maps S 2 S 2 .
Hint. If Rnk is uniformly close to F : S 2 S 2 , the maps must have the same degree, as
shown in basic topology courses.
We say a point S 2 belongs to the Fatou set of R provided there exists a neighborhood
of such that {Rn : n 1} is a normal family, with respect to (, S 2 ). The Fatou
set of R is denoted FR .

4. Show that FR is open, R : FR FR , and {Rn F : n 1} is normal with respect to

R
(FR , S 2 ).
The complement of the Fatou set is called the Julia set, JR = S 2 \ FR . By Exercise 3,
JR 6= , whenever deg R 2, which we assume from here on.
5. Given JR , and any neighborhood O of in S 2 , consider
[
(27.5)
EO = S 2 \
Rn (O).
n0

Use Theorem 27.1 to show that EO contains at most 2 points.

6. Set
(27.6)

E =

{EO : O neighborhood of }.

Show that E = EO for some neighborhood O of . Show that R : E E .

7. Consider the function Sq : S 2 S 2 , given by Sq(z) = z 2 , Sq() = . Show that
(27.7)

JSq = {z : |z| = 1},

E = {0, },

JSq .

188

8. Suppose E consists of one point. Show that R is conjugate to a polynomial, i.e., there
exists a linear fractional transformation L such that L1 RL is a polynomial.
Hint. Consider the case E = {}.
9. Suppose E consists of two points. Show that R is conjugate to Pm for some m Z,
where Pm (z) = z m , defined appropriately at 0 and .
Hint. Suppose E = {0, }. Then R either fixes 0 and or interchanges them.
Conclusion. Typically, E = , for JR . Furthermore, if E 6= , then E = E is
independent of JR , and E FR .
10. Show that
(27.8)

R : JR JR

and
R1 (JR ) JR .

(27.9)

Hint. Use Exercise 4. For (27.8), if R() FR , then has a neighborhood O such that
R(O) FR . For (27.9), use R : FR FR .
11. Show that either JR = S 2 or JR has empty interior.
Hint. If JR has a neighborhood O JR , then, by (27.8), Rn (O) JR . Now use
Exercise 5.
12. Show that, if p JR , then
(27.10)

Rk (p) is dense in JR .

k0

Hint. Use Exercises 56, the conclusion following Exercise 9, and Exercise 10.
13. Show that
(27.11)

R(z) = 1

2
= JR = S 2 .
z2

Remark. This could be tough. See [CG], p. 82, for a proof of (27.11), using results not
developed in these exercises.
14. Show that
(27.12)

R(z) = z 2 2 = JR = [2, 2] C.

Hint. Show that R : [2, 2] [2, 2], and that if z0 C \ [2, 2], then Rk (z0 ) as
k .
The next exercise will exploit the following general result.

189

Proposition 27.3. Let X be a compact metric space, F : X X a continuous map.

Assume that for each nonempty open U X, there exists N (U ) N such that
[

F j (U ) = X.

0jN (U )

Then there exists p X such that

[

F j (p) is dense in X.

j1

15. Show that Proposition 27.3 applies to R : JR JR .

Hint. Use Exercise 5, and the conclusion after Exercise 9.
16. Show that, for R : S 2 S 2 as above,
p JR such that

Rj (p) is dense in JR .

j1

17. Prove Proposition 27.3.

Hint. Take a countable dense subset {qj : j 1} of X. Try to produce a shrinking family
Kj Kj+1 of nonempty, compact subsets of X, and Nj N, such that, for all
j N,
F Nj (Kj ) B2j (qj ).
Then take p j1 Kj , so
F Nj (p) B2j (qj ),

j 1.

18. Show that R : JR JR is surjective.

Hint. Consider Exercise 15.
19. Show that, for each k N, JRk = JR .
Hint. Clearly FRk FR . To get the converse, use Rj = R` Rmk , 0 ` k 1.
20. Show that JR must be infinite. (Recall that we assume deg R 2.)
e and find p JR such that R(p)
e
Hint. If JR is finite, we can replace R by Rk = R
= p.
Then take a small neighborhood O of p (disjoint from the rest of JR ) and apply Exercise
e to get a contradiction.
5 (and Exercise 15) to R,
21. Show that JR has no isolated points.
Hint. If p JR is isolated, let O be a small neighborhood of p in S 2 , disjoint from the

190

rest of JR , and (again) apply Exercise 5 (and Exercise 15) to R, to get a contradiction,
taking into account Exercise 20.

These exercises provide a glimpse at an area known as Complex Dynamics. More material
on this area can be found in Chapter 5 of [Sch] (a brief treatment), and in [CG] and [Mil].
As opposed to (27.7) and (27.12), typically JR has an extremely complicated, fractal
structure, as explained in these references.

191

28. Picards theorems

Here we establish two theorems of E. Picard. The first, known as Picards little
theorem, is an immediate consequence of the fact that the disk holomorphically covers
C \ {0, 1}.
Proposition 28.1. If p and q are distinct points in C and if f : C C \ {p, q} is
holomorphic, then it is constant.
Proof. Without loss of generality, we can take p = 0, q = 1. Via the covering map : D
C \ {0, 1} produced in 25, f lifts to a holomorphic map
(28.1)

g : C D,

f = g.

Liouvilles theorem then implies g is constant, so also f is constant.

The following sharper result is called Picards big theorem. It is proved using Montels
theorem.
Proposition 28.2. If p and q are distinct and
(28.2)

f : D \ {0} C \ {p, q}

is holomorphic, then the singularity at 0 is either a pole or a removable singularity. Equivalently, if f : D \ {0} S 2 is holomorphic and has range in S 2 \ {p, q, r} with p, q, r S 2
distinct, then f extends to f : D S 2 , holomorphic.
Proof. Again there is no loss of generality in taking p = 0, q = 1, and r = . Take
= D \ {0} and define f : C \ {0, 1} by f (z) = f (2 z). By Montels theorem
(Theorem 27.1), {f } is a normal family of maps from to S 2 . In particular, there exists
a subsequence fk converging locally uniformly on :
fk g : S 2 ,

(28.3)

and g is a holomorphic map. Pick r (0, 1) and set = {z : |z| = r} . The

convergence in (28.3) is uniform on .
We consider two cases.
Case A.
/ g().
Then there exists A < such that
(28.4)

|fk (z)| A,

|z| = r,

for large k, or equivalently

(28.5)

|f (z)| A,

|z| = 2k r.

192

The maximum principle then implies |f (z)| A for all z close to 0, so 0 is a removable
singularity of f .
Case B. g().
By Corollary 27.2, g on . Consequently, 1/fk 0 uniformly on , and then
analogues of (28.4)(28.5) hold with f replaced by 1/f (recall that f is nowhere vanishing
on D \ {0}), so f has a pole at 0.

Exercises
1. Fix distinct points a, b C, let be the line segment joining a and b, and assume
f : C C \ is holomorphic. Show that f is constant. Show this in an elementary
fashion, not using Picard theorems.
2. Suppose f : C C is a holomorphic function that is not a polynomial. (We say f
is transcendental.) Show that for every w C, with at most one exception, there are
infinitely many z C such that f (z ) = w.
3. Suppose that f is meromorphic on C. Show that f can omit at most two complex
values. Give an example where such an omission occurs.

193

29. Harmonic functions again: Harnack estimates and more Liouville theorems
Here we study further properties of harmonic functions. A key tool will be the explicit
formula for the solution to the Dirichlet problem

(29.1)
u = 0 on D1 (0), uS 1 = f,
given in 13, namely
Z
(29.2)

p(z, )f () d,

u(z) =
0

where
(29.3)

p(z, ) =

1 1 |z|2
,
2 |w z|2

w = ei ,

as in (13.24), or equivalently
(29.4)

p(z, ) =

1
ei + z
1
ei
1
Re i
= Re i

.
2
e z

e z
2

We restrict attention to harmonic functions on domains in C = R2 . However we point

out that results presented here can be extended to treat harmonic functions on domains
in Rn . In such a case we replace (29.2)(29.3) by
Z
(29.5)
u(x) =
p(x, )f () dS(),
S n1

(29.6)

p(x, ) =

1 |x|2
,
An1 |x |n
1

where An1 is the (n 1)-dimensional area of the unit sphere S n1 Rn . A proof of

(29.5)(29.6) can be found in Chapter 5 of [T2]. But here we will say no more about the
higher-dimensional case.
We use (29.2)(29.3) to establish a Harnack estimate, concerning the set of harmonic
functions
(29.7)

194

Proposition 29.1. For z D,

(29.8)

u A = u(z)

1 |z|
.
1 + |z|

Proof. The mean value property of harmonic functions implies that f = u|D satisfies
Z 2
1
(29.9)
f () d = 1,
2 0
if u A. Given this, (29.8) follows from (29.2)(29.3) together with the estimate
(29.10)

1 |z|2
1 |z|
,

|w z|2
1 + |z|

To get (29.10), note that if w = ei , z = rei , and = , then

|w z|2 = (ei r)(ei r)
= 1 2r cos + r2

(29.11)

1 + 2r + r2 = (1 + r)2 ,
from which (29.10) follows.
By translating and scaling, we deduce that if u is 0 and harmonic on DR (p), then
(29.12)

|z p| = a [0, R) = u(z)

Ra
u(p).
R+a

This leads to the following extension of the Liouville theorem, Proposition 7.5 (with a
different approach from that suggested in Exercise 1 of 7).
Proposition 29.2. If u is harmonic on all of C (and real valued) and bounded from below,
then u is constant.
Proof. Adding a constant if necessary, we can assume u 0 on C. Pick z0 , z1 C and
set a = |z0 z1 |. By (29.12),
(29.13)

u(z0 )

Ra
u(z1 ),
R+a

R (a, ),

hence, taking R ,
(29.14)

u(z0 ) u(z1 ),

(29.15)

u(z0 ) = u(z1 ),

and proves the proposition.

The following result will lead to further extensions of Liouvilles theorem.

195

Proposition 29.3. There exists a number A (0, ) with the following property. Let u
be harmonic in DR (0). Assume
(29.16)

u(0) = 0,

u(z) M

on DR (0).

Then
(29.17)

(29.18)

p DR/2 (0),

u(p) =

inf

DR/2 (0)

u.

From (29.12), with R replaced by R/2, a by R/4, and u by v, we get

(29.19)

v(z)

1
M u(p) on DR/2 (p).
3

Hence
(29.20)

1
Area DR (0)

ZZ
v(z) dx dy

1 1
M u(p) .
16 3

DR (0)

On the other hand, the mean value property for harmonic functions implies that the left
side of (29.20) equals v(0) = M , so we get
(29.21)

M u(p) 48M,

which implies (29.17).

Note that Proposition 29.3 gives a second proof of Proposition 29.2. Namely, under
the hypotheses of Proposition 29.2, if we set v(z) = u(0) u(z), we have v(0) = 0 and
v(z) u(0) on C (if u 0 on C), hence, by Proposition 29.3, v(z) Au(0) on C, so v
is bounded on C, and Proposition 7.5 implies v is constant. Note however, that the first
proof of Proposition 29.2 did not depend upon Proposition 7.5.
Here is another corollary of Proposition 29.3.
Proposition 29.4. Assume u is harmonic on C and there exist C0 , C1 (0, ), and
k Z+ such that
(29.22)

u(z) C0 + C1 |z|k ,

z C.

Then there exist C2 , C3 (0, ) such that

(29.23)

u(z) C2 C3 |z|k ,

z C.

196

Proof. Apply Proposition 29.3 to u(z) u(0), M = C0 + |u(0)| + C1 Rk .

Note that as long as C2 C0 and C3 C1 , the two one-sided bounds (29.22) and
(29.23) imply
(29.24)

|u(z)| C2 + C3 |z|k ,

z C.

We aim to show that if u(z) is harmonic on C and satisfies the bound (29.24), then u
must be a polynomial in x and y. For this, it is useful to have estimates on derivatives
xi yj u, which we turn to.
If u C 2 (D) C(D) is harmonic on D = D1 (0), the formula (25.2) yields
Z
(29.25)

xi yj u(z)

pij (z, )f () d,
0

where
pij (z, ) = xi yj p(z, ).

(29.26)

From (29.3) it is apparent that p(z, ) is smooth in z D. We have bounds of the form
(29.27)

|z|

1
.
2

(29.28)

1
ei
p(z, ) = Re i
,
x

(e z)2

1
iei
p(z, ) = Re i
.
y

(e z)2

Hence
1
1
|ei z|2
1

,
(1 |z|)2

|x,y p(z, )|
(29.29)

the last estimate by a variant of (29.11).

Applied to (29.25), the bounds (29.27) imply
(29.30)

|z|1/2

whenever u C 2 (D) C(D) is harmonic on D.

We are now are ready for the following.

197

Proposition 29.5. If u : C R is harmonic and, for some k Z+ , Cj (0, ),

(29.31)

u(z) C0 + C1 |z|k ,

z C,

then u is a polynomial in x and y of degree k.

Proof. By Proposition 29.4, we have the two-sided bound
(29.32)

|u(z)| C2 + C3 |z|k ,

z C.

Now set vR (z) = Rk u(Rz). We have vR |D bounded, independent of R [1, ), where

D = D1 (0). Hence, by (29.30), xi yj vR is bounded on D1/2 (0), independent of R, for each
i, j Z+ , so
(29.33)

|z|

1
, R [1, ).
2

Taking R yields xi yj u = 0 on C for i + j > k, which proves the proposition.

Remark. In case k = 0, the argument above gives another proof of Proposition 7.5.

Exercises
1. Suppose f : C C is holomorphic and |f (z)| AeB|z| . Show that if f has only finitely
many zeros then f has the form
f (z) = p(z) eaz+b ,
for some polynomial p(z).
Hint. If p(z) has the same zeros as f (z), write f (z)/p(z) = eg(z) and apply Proposition
29.5 to Re g(z).
2. Show that the function ez z (considered in Exercise 10 of 6) has infinitely many
zeros. More generally, show that, for each a C, ez z a has infinitely many zeros.
3. The proof given above of Proposition 29.3 shows that (29.17) holds with A = 47. Can
you show it holds with some smaller A?

198

30. Periodic and doubly periodic functions - infinite series representations

We can obtain periodic meromorphic functions by summing translates of z k . For
example,
(30.1)

f1 (z) =

1
(z n)2
n=

is meromorphic on C, with poles in Z, and satisfies f1 (z + 1) = f1 (z). In fact, we have

(30.2)

1
2
=
.
2
2
(z

n)
sin
z
n=

To see this, note that both sides have the same poles, and their difference g1 (z) is seen to
be an entire function, satisfying g1 (z + 1) = g1 (z). Also it is seen that, for z = x + iy, both
sides of (30.2) tend to 0 as |y| . This forces g1 0.
A second example is
1
1 X 1
1
f2 (z) = lim
= +
+
m
zn
z
zn n
n=m
m
X

(30.3)

n6=0

1
2z
+
.
2
z n=1 z n2

This is also meromorphic on C, with poles in Z, and it is seen to satisfy f2 (z + 1) = f2 (z).

We claim that
1 X 1
1
(30.4)
+
+
= cot z.
z
zn n
n6=0

In this case again we see that the difference g2 (z) is entire. Furthermore, applying d/dz
to both sides of (30.4), we get the two sides of (30.2), so g2 is constant. Looking at the
last term in (30.3), we see that the left side of (30.4) is odd in z; so is the right side; hence
g2 = 0.
As a third example, we consider
m
1
X
(1)n
1 X
1
n
lim
= +
(1)
+
m
zn
z
zn n
n=m

(30.5)

n6=0

1
2z
+
(1)n 2
z n=1
z n2

X
1
z(1 2k)
= 4
.
2
z
[z (2k 1)2 ][z 2 (2k)2 ]
k=1

199

We claim that
1
1 X
1

+
(1)n
+
=
.
z
zn n
sin z

(30.6)

n6=0

In this case we see that their difference g3 (z) is entire and satisfies g3 (z + 2) = g3 (z). Also,
for z = x + iy, both sides of (30.6) tend to 0 as |y| , so g3 0.
We now use a similar device to construct doubly periodic meromorphic functions, following K. Weierstrass. These functions are also called elliptic functions. Further introductory
material on this topic can be found in [Ahl] and [Hil]. Pick 1 , 2 C, linearly independent
over R, and form the lattice
(30.7)

= {j1 + k2 : j, k Z}.

(30.8)

(z; ) =

X
1
1
1
+

.
z2
(z )2
2
06=

Convergence on C \ is a consequence of the estimate

(30.9)

1
|z|
1
2 C 3,
2
(z )

||

for || 2|z|.

To verify that
(30.10)

(z + ; ) = (z; ),

(30.11)

0 (z; ) = 2

1
,
(z )3

which clearly satisfies

(30.12)

0 (z + ; ) = 0 (z; ),

Hence
(30.13)

(z + ; ) (z; ) = c(),

Now (30.8) implies (z; ) = (z; ). Hence, taking z = /2 in (30.13) gives c() = 0
for all , and we have (30.10).

200

Another analogy with (30.4) leads us to look at the function (not to be confused with
the Riemann zeta function)
(30.14)

(z; ) =

X 1
1
1
z
+
+ + 2 .
z
z
06=

We note that the sum here is obtained from the sum in (30.8) (up to sign) by integrating
from 0 to z along any path that avoids the poles. This is enough to establish convergence
of (30.14) in C \ , and we have
0 (z; ) = (z; ).

(30.15)
In view of (30.10), we hence have
(30.16)

(z + ; ) (z; ) = (),

In this case () 6= 0, but we can take a, b C and form

(30.17)

a,b (z; ) = (z a; ) (z b; ),

obtaining a meromorphic function with poles at (a + ) (b + ), all simple (if a b

/ ).
Let us compare the doubly periodic function constructed in (24.8)(24.11), which
maps the rectangle with vertices at 1, 1, 1 + ip, 1 + ip conformally onto the upper half
plane U , with (1) = 1, (0) = 0, (1) = 1. (Here p is a given positive number.) As
seen there,
(30.18)

(z + ) = (z),

= {4k + 2i`p : k, ` Z}.

Furthermore, this function has simple poles at (ip + ) (ip + 2 + ), and the residues at
ip and at ip + 2 cancel. Thus there exist constants A and B such that
(30.19)

(z) = Aip,ip+2 (z; ) + B.

The constants A and B can be evaluated by taking z = 0, 1, though the resulting formulas
give A and B in terms of special values of (z; ) rather than in elementary terms.

Exercises
1. Setting z = 1/2 in (30.2), show that

X
1
2
=
.
2
n
6
n=1

201

Compare (13.79). Differentiate (30.2) repeatedly and obtain formulas for

even integers k.
Hint. Denoting the right side of (30.2) by f (z), show that
(`)

P
n1

nk for

(z n)(`+2) .

n=

f
Meanwhile, use

(2k2)

1
2

2k

P
n=1

n2k .

n1,odd

n=1

= (2k 1)!22k+1

2k

+2

2k

n2k

n=1

n1,odd

n2k , in terms of f (2k2) (1/2).

1A. Set F (z) = ( cot z) 1/z, and use (30.4) to compute F (`) (0). Show that, for |z| < 1,

cot z =

X
1
2
(2k)z 2k1 ,
z

(2k) =

k=1

n2k .

n=1

1B. Recall from Exercise 6 in 12 that, for |z| sufficiently small,

1 ez + 1
1 X
Bk 2k1
=
+
(1)k1
z
,
z
2 e 1
z
(2k)!
k=1

with Bk (called the Bernoulli numbers) rational numbers for each k. Note that
e2iz + 1
1
= cot z.
2iz
e
1
i
Deduce from this and Exercise 1A that, for k 1,
2(2k) = (2)2k

Bk
.
(2k)!

Relate this to results of Exercise 1.

1C. For an alternative aproach to the results of Exercise 1B, show that
G(z) = cot z = G0 (z) = 2 G(z)2 .

202

Using

G(z) =

1 X
+
an z 2n1 ,
z n=1

compute the Laurent series expansions of G0 (z) and G(z)2 and deduce that a1 = 2 /3,
while, for n 2,
n1
1 X
an =
an` a` .
2n + 1
`=1

In concert with Exercise 1A, show that (2) = 2 /6, (4) = 4 /90, and also compute (6)
and (8).
2. Set
F (z) = z

Y
n=1

Show that

z2
1 2 .
n

F 0 (z)
1 X 2z
= +
.
F (z)
z n=1 z 2 n2

Using this and (30.3)(30.4), deduce that

F (z) = sin z,
obtaining another proof of (18.21).
Hint. Show that if F and G are meromorphic and F 0 /F G0 /G, then F = cG for some
constant c. To find c in this case, note that F 0 (0) = .
3. Show that if is a lattice of the form (30.7) then a meromorphic function satisfying
(30.20)

f (z + ) = f (z),

yields a meromorphic function on the torus T , defined by (26.14). Show that if such f
has no poles then it must be constant.
We say a parallelogram P C is a period parallelogram for a lattice (of the form (30.7))
provided it has vertices of the form p, p + 1 , p + 2 , p + 1 + 2 . Given a meromorphic
function f satisfying (30.20), pick a period parallelogram P whose boundary is disjoint
from the set of poles of f .
4. Show that

Z
f (z) dz = 0.
P

203

Deduce that

Respj (f ) = 0.

pj P

Deduce that if f has just one pole in P then that pole cannot be simple.
5. For defined by (30.14), show that, if Im(2 /1 ) > 0,
Z
(30.21)
(z; ) dz = (1 )2 (2 )1 = 2i.
P

6. Show that in (30.16) satisfies

(30.22)

( + 0 ) = () + ( 0 ),

, 0 .

Show that if , /2
/ , then
() = 2(/2; ).
7. Apply Greens theorem
ZZ
Z
g
f

dx dy = (f dx + g dy)
x y

in concert with 0 (z; ) = (z; ), written as

1
1
+
(z; ) = (z; ),
2 x
i y
and with = P, as in Exercise 5, to establish that
(30.23)

(1 ) 2 (2 ) 1 = 2iI(),

with
ZZ
(30.24)

I() = lim

0
P\D (0)

(z; ) dx dy,

assuming P is centered at 0.
8. Solve the pair of equations (30.21) and (30.23) for (1 ) and (2 ). Use this in
concert with (30.22) to show that
(30.25)

() =

1
I() + ,
A(P)

204

where I() is as in (30.24) and A(P) is the area of P.

9. Show that the constant A in (30.19) satisfies
A = Resip ().
10. Show that the constants A and B in (30.19) satisfy
[(1 ip; ) (1 ip; )]A + B = 1,
and
(4)A + 2B = 0,
with given by (30.18).
Hint. Both (z) and (z; ) are odd in z.
In Exercises 1112, given pj T , nj Z+ , set =
(30.26)
Set || =

nj pj and define

M (T ) = {f meromorphic on T : poles of f are at pj and of order nj }.

P

nj .

11. Show that || = 2 dim M (T ) = 2, and that this space is spanned by 1 and p1 ,p2
if n1 = n2 = 1, and by 1 and (z p1 ) if n1 = 2.
Hint. Use Exercise 4.
12. Show that
(30.27)

|| = k 2 = dim M (T ) = k.

Hint. Argue by induction on k, noting that you can augment || by 1 either by adding
another pj or by increasing some positive nj by 1.

205

31. The Weierstrass in elliptic function theory

It turns out that a general elliptic function with period lattice can be expressed in
terms of (z; ) and its first derivative. Before discussing a general result, we illustrate
this in the case of the functions a,b (z; ), given by (30.17). Henceforth we simply denote
these functions by (z) and a,b (z), respectively.
We claim that, if 2
/ , then
0 ()
= , (z) + 2().
(z) ()

(31.1)

To see this, note that both sides have simple poles at z = . (As will be shown below, the
zeros of 0 (z) satisfy 2 .) The factor 0 () makes the poles cancel, so the difference
is entire, hence constant. Both sides vanish at z = 0, so this constant is zero. We also note
that
(31.2)

a,b (z) = , (z ),

a+b
ab
, =
.
2
2

As long as a b
/ , (31.1) applies, giving
(31.3)

a,b (z) =

0 ()
2(),
(z ) ()

a+b
ab
, =
.
2
2

We now prove the result on the zeros of 0 (z) stated above. Assume has the form
(30.7).
Proposition 31.1. The three points 1 /2, 2 /2 and (1 +2 )/2 are (mod ) all the zeros
of 0 (z).
Proof. Symmetry considerations (oddness of 0 (z)) imply 0 (z) = 0 at each of these three
points. Since 0 (z) has a single pole of order 3 in a period parallelogram, these must be
all the zeros. (Cf. Exercise 1 below to justify this last point.)
The general result hinted at above is the following.
Proposition 31.2. Let f be an elliptic function with period lattice . There exist rational
functions Q and R such that
(31.4)

f (z) = Q((z)) + R((z))0 (z).

Proof. First assume f is even, i.e., f (z) = f (z). The product of f (z) with factors of
the form (z) (a) lowers the degree of a pole of f at any point a
/ , so there exists
a polynomial P such that g(z) = P ((z))f (z) has poles only in . Note that g(z) is

206

also even. Then there exists a polynomial P2 such that g(z) P2 ((z)) has its poles
annihilated. This function must hence be constant. Hence any even elliptic f must be a
rational function of (z).
On the other hand, if f (z) is odd, then f (z)/0 (z) is even, and the previous argument
applies, so a general elliptic function must have the form (31.4).
The right side of (31.3) does not have the form (31.4), but we can come closer to this
form via the identity
(31.5)

(z ) = (z) () +

1 0 (z) + 0 () 2
.
4 (z) ()

This identity can be verified by showing that the difference of the two sides is pole free and
vanishes at z = 0. The right side of (31.5) has the form (31.4) except for the occurrence
of 0 (z)2 , which we will dispose of shortly.
Note that 0 (z)2 is even, with poles (of order 6) on . We can explicitly write this as
P ((z)), as follows. Set
(31.6)

ej =

j
,
2

j = 1, 2, 3,

(31.7)

0 (z)2 = 4((z) e1 )((z) e2 )((z) e3 ).

In fact, both sides of (31.7) have poles of order 6, precisely at points of . Furthermore,
by Proposition 31.1, the zeros of 0 (z)2 occur precisely at z = j (mod ), each zero
having multiplicity 2. We also see that the right side of (31.7) has a double zero at
z = j , j = 1, 2, 3. So the quotient is entire, hence constant. The factor 4 arises by
examining the behavior as z 0.
The identity (31.7) is a differential equation for (z). Separation of variables yields
(31.8)

1
2

( e1 )( e2 )( e3 )

= z + c.

The left side of (31.8) is known as an elliptic integral.

Any cubic polynomial in u is a constant multiple of (u e1 )(u e2 )(u e3 ) for some
ej C. However, it is not quite the case that every cubic polynomial fits into the current
setting. Here is one constraint; another will be produced in (31.15) below.
Proposition 31.2. Given a lattice C, the quantities ej in (31.6) are all distinct.
Proof. Note that (z) ej has a double pole at each z , and a double zero at z = j /2.
Hence, in an appropriate period parallelogram, it has no other zeros (again cf. Exercise 1
below). Hence (k /2) ej = ek ej 6= 0 for j 6= k.

207

We can get more insight into the differential equation (31.7) by comparing Laurent
series expansions of the two sides about z = 0. First, we can deduce from (30.8) that
(31.9)

(z) =

1
+ az 2 + bz 4 + .
2
z

Of course, only even powers of z arise. Regarding the absence of the constant term and
the values of a and b, see Exercise 3 below. We have
(31.10)

a=3

X
06=

1
,
4

b=5

06=

1
.
6

Hence
(31.11)

0 (z) =

2
+ 2az + 4bz 3 + .
z3

It follows, after some computation, that

(31.12)

1 0 2
1
2a
(z) = 6 2 4b + ,
4
z
z

while
((z) e1 )((z) e2 )((z) e3 )
(31.13)

= (z)3 1 (z)2 + 2 (z) 3

1
1
3a + 2
= 6 4+
+ (3b 2a1 3 ) + ,
z
z
z2

where
1 = e1 + e2 + e3 ,
(31.14)

2 = e1 e2 + e2 e3 + e3 e1 ,
3 = e1 e2 e3 .

Comparing coefficients in (31.12)(31.13) gives the following relation:

(31.15)

e1 + e2 + e3 = 0.

It also gives
(31.16)

e1 e2 + e2 e3 + e1 e3 = 5a,
e1 e2 e3 = 7b,

where a and b are given by (31.10). Hence we can rewrite the differential equation (31.7)
as
(31.17)

0 (z)2 = 4(z)3 g2 (z) g3 ,

208

where g2 and g3 , known as the Weierstrass invariants of the lattice , are given by
(31.18)

g2 = 60

06=

1
,
4

g3 = 140

X
06=

1
.
6

Exercises
1. Assume f is meromorphic (and not identically zero) on T = C/. Show that the
number of poles of f is equal to the number of zeros of f , counting multiplicity.
Hint. Let bound a period parallelogram, avoiding the zeros and poles of f , and examine
1
2i

f 0 (z)
dz.
f (z)

Recall the argument principle, discussed in 17.

2. Show that, given a lattice C, and given C,
(31.19)

+ z; =
z; ,
2
2

z.

3. Consider
(31.20)

X
1
1
1
(z) = (z) 2 =
2 ,
z
(z )2

\0

(31.21)

X
1 (k)
(0) = (k + 1)
(k+2) .
k!
\0

(These quantities vanish for k odd.) Relate these results to (31.9)(31.10).

4. Complete the sketch of the proof of (31.5).
Hint. Use the fact that (z) z 2 is holomorphic near z = 0 and vanishes at z = 0.
5. Deduce from (31.17) that
(31.22)

1
00 (z) = 6(z)2 g2 .
2

209

6. Say that, near z = 0,

(31.23)

(z) =

X
1
+
bn z 2n ,
2
z
n1

where bn are given by (31.21), with k = 2n. Deduce from Exercise 5 that for n 3,
(31.24)

bn =

n2
X
3
bk bnk1 .
(2n + 3)(n 2)
k=1

In particular, we have
b3 =
and
b5 =

1 2
b ,
3 1

b4 =

3
b1 b2 ,
11

1 2
1 2 2 3
(b2 + 2b1 b3 ) =
b + b .
13
13 2 3 1

(31.25)

n =

X
\0

1
,
2n

then for n 3,
(31.26)

n = Pn (2 , 3 ),

where Pn (2 , 3 ) is a polynomial in 2 and 3 with coefficients that are positive, rational

numbers. Use (31.16) to show that
(31.27)

2 =

1
(e1 e2 + e2 e3 + e1 e3 ),
15

3 =

1
e1 e2 e3 .
35

Note that bn = (2n + 1)n+1 . Note also that gk in (31.17)(31.18) satisfy g2 = 602 and
g3 = 1403 .
8. Given f as in Exercise 1, show that
(31.28)

1
2i

f 0 (z)
dz Z,
f (z)

210

9. Again take f as in Exercise 1. Assume f has zeros at pj T , of order mj , and poles

at qj T , of order nj , and no other zeros or poles. Show that
X
X
(31.29)
mj pj
nj qj = 0 (mod ).
Hint. Take as in Exercise 1, and consider
Z 0
f (z)
1
z dz.
(31.30)
2i
f (z)

On the one hand, Cauchys integral theorem (compare (5.19)) implies (31.30) is equal
to the left side of (31.29), provided pj and qj are all in the period domain. On the other
hand, if consists of four consecutive line segments, 1 , 2 , 3 , 4 , periodicity of f 0 (z)/f (z)
implies that (31.30) equals
Z 0
Z 0
2
1
f (z)
f (z)
(31.31)

dz +
dz.
2i
f (z)
2i
f (z)
1

Use Exercise 8 to deduce that the coefficients of 1 and 2 in (31.31) are integers.
10. Deduce from (31.5) that

(u) 0 (u) 1
u + v + w = 0 det (v) 0 (v) 1 = 0.
(w) 0 (w) 1

(31.32)

(31.33)

(2z) =

1 00 (z) 2
2(z).
4 0 (z)

Hint. Set = z + h in (31.5) and let h 0.

12. Deduce from (31.33), in concert with (31.17) and (31.22), that
(2z) = R((z)),
with
R() =

4 + (g2 /2) 2 + 2g3 + (g2 /4)2

.
4 3 g2 g3

13. Use (31.3) and (31.5), plus (31.7), to write a,b (z) (as in (31.3)) in the form (31.4), i.e.,
a,b (z) = Q((z)) + R((z))0 (z).

211

32. Theta functions and

We begin with the function
(32.1)

(x, t) =

en t e2inx ,

nZ

1 2
=
.
t
4 x2

(32.2)

Note that is actually holomorphic on {(x, t) C C : Re t > 0}. It is periodic of period

1 in x; (x + 1, t) = (x, t). Also one has
(x + it, t) = et2ix (x, t).

(32.3)

This identity will ultimately lead us to a connection with (z). In addition, we have
2
1 X
(1)n en t e2inx ,
x + ,t =
2

(32.4)

nZ

and
(32.5)

X
2
i
x + t, t = et/4
e(n+1/2) t e2inx ,
2
nZ

which introduces series related to but slightly different from that in (32.1).
Following standard terminology, we set t = i , with Im > 0, and denote (z, i )
by
(32.6)

3 (z, ) =

en

i 2niz

nZ

p2n q n ,

nZ

where
p = eiz ,

(32.7)

q = ei .

(32.8)

4 (z, ) =

(1)n en

nZ

i 2inz

(1)n p2n q n ,

nZ

212

and
(32.9)

1 (z, ) = i

(1)n e(n1/2)

i (2n1)iz

=i

nZ

and finally
(32.10)

2 (z, ) =

(1)n p2n1 q (n1/2) ,

nZ

e(n1/2)

i (2n1)iz

nZ

p2n1 q (n1/2) .

nZ

We will explore these functions, with the goal of relating them to (z). These results are
due to Jacobi; we follow the exposition of [MM].
To begin, we record how j (z + ) is related to k (z) for various values of . Here and
(usually) below we will suppress the and denote j (z, ) by j (z), for short. In the table
below we use
a = p1 q 1/4 = eizi /4 ,

b = p2 q 1 = e2izi .

1
2
3
4

z + 1/2

z + /2

z + 1/2 + /2

z+1

z+

z+1+

2
1
4
3

ia4
a3
a2
ia1

a3
ia4
ia1
a2

1
2
3
4

b1
b2
b3
b4

b1
b2
b3
b4

(z) 2
j
(32.11)
Fjk (z) =
k (z)
satisfy
(32.12)

Fjk (z + ) = Fjk (z),

where
(32.13)

= {k + ` : k, ` Z}.

Note also that

(32.14)

Gj (z) =

0j (z)
j (z)

satisfies
(32.15)

Gj (z + 1) = Gj (z),

Gj (z + ) = Gj (z) 2i.

To relate the functions Fjk to previously studied elliptic functions, we need to know the
zeros of k (z). Here is the statement:

213

Proposition 32.1. We have

(32.16)

1
2 (z) = 0 z + ,
2
1

3 (z) = 0 z + + , 4 (z) = 0 z + .
2 2
2
1 (z) = 0 z ,

Proof. In view of the relations tabulated above, it suffices to treat 1 (z). We first note
that
(32.17)

1 (z) = 1 (z).

To see this, replace z by z in (32.8) and simultaneously replace n by m. Then replace

m by n 1 and observe that (32.17) pops out. Hence 1 has a zero at z = 0. We claim it
is simple and that 1 has no others, mod . To see this, let be the boundary of a period
parallelogram containing 0 in its interior. Then use of (32.15) with j = 1 easily gives
Z 0
1
1 (z)
dz = 1,
2i
1 (z)

completing the proof.

Let us record the following complement to (32.17):
(32.18)

2 j 4 = j (z) = j (z).

The proof is straightforward from the defining formulas (32.6)(32.9).

We are now ready for the following important result. For consistency with [MM], we
slightly reorder the quantities e1 , e2 , e3 . Instead of using (31.6), we set

+

1
1
2
2
(32.19)
e1 =
, e2 =
, e3 =
,
2
2
2
where, in the current setting, with given by (32.13), we take 1 = 1 and 2 = .
Proposition 32.2. For (z) = (z; ), with of the form (32.13) and j (z) = j (z, ),

(32.20)

0 (0) (z) 2
2
1

(z) = e1 +
1 (z) 2 (0)
0 (0) (z) 2
3
1
= e2 +

1 (z) 3 (0)
0 (0) (z) 2
4
1
= e3 +

.
1 (z) 4 (0)

Proof. We have from (32.11)(32.13) that each function Pj (z) on the right side of (32.20)
is -periodic. Also Proposition 32.1 implies each Pj has poles of order 2, precisely on

214

. Furthermore, we have arranged that each such function has leading singularity z 2
as z 0, and each Pj is even, by (32.17) and (32.18), so the difference (z) Pj (z) is
constant for each j. Evaluating at z = 1/2, (1 + )/2, and /2, respectively, shows that
these constants are zero, and completes the proof.
Part of the interest in (32.20) is that the series (32.6)(32.10) for the theta functions are
extremely rapidly convergent. To complete this result, we want to express the quantities
ej in terms of theta functions. The following is a key step.
Proposition 32.3. In the setting of Proposition 32.2,
0 (0) (0) 2
4
1
= 2 4 (0)4 ,
2 (0)3 (0)
0 (0) (0) 2
3
1
e1 e3 =
= 2 3 (0)4 ,
2 (0)4 (0)
0 (0) (0) 2
2
1
e2 e3 =
= 2 2 (0)4 .
3 (0)4 (0)
e1 e2 =

(32.21)

Proof. To get the first part of the first line, evaluate the second identity in (32.20) at
z = 1/2, to obtain
0 (0) (1/2) 2
3
1

,
e1 e2 =
1 (1/2) 3 (0)
and then consult the table to rewrite 3 (1/2)/1 (1/2). Similar arguments give the first
identity in the second and third lines of (32.21). The proof of the rest of the identities
then follows from the next result.
Proposition 32.4. We have
(32.22)

Proof. To begin, consider

(z) = 1 (2z)1 1 (z)2 (z)3 (z)4 (z).
Consultation of the table shows that (z + ) = (z) for each . Also is free of
poles, so it is constant. The behavior as z 0 reveals the constant, and yields the identity
(32.23)

1 (2z) = 2

.
2 (0)3 (0)4 (0)

(32.24)

215

(a consequence of (32.17)(32.18)), yields

000
002 (0) 003 (0) 004 (0)
1 (0)
=
+
+
.
01 (0)
2 (0)
3 (0)
4 (0)

(32.25)
Now, from (32.2) we have

j
1 2 j
=
,

4i z 2

(32.26)
and computing
(32.27)

i
h
log 2 (0, ) + log 3 (0, ) + log 4 (0, ) log 01 (0, )

(32.28)

2 (0, )3 (0, )4 (0, )/01 (0, ) is independent of .

Thus
01 (0) = A2 (0)3 (0)4 (0),

(32.29)

(32.30)

01 (0) 2q 1/4 ,

2 (0) 2q 1/4 ,

3 (0) 1,

4 (0) 1,

which implies A = , proving (32.22).

Now that we have Proposition 32.3, we can use (e1 e3 ) (e1 e2 ) = e2 e3 to deduce
that
(32.31)

(32.32)

e1 + e2 + e3 = 0

to deduce the following.

Proposition 32.5. In the setting of Proposition 32.2, we have

2
3 (0)4 + 4 (0)4 ,
3

2
e2 =
2 (0)4 4 (0)4 ,
3

2
e3 =
2 (0)4 + 3 (0)4 .
3

e1 =
(32.33)

216

Thus we have an efficient method to compute (z; ) when has the form (32.13). To
pass to the general case, we can use the identity
1 z
(32.34)
(z; a) = 2 ; .
a
a
See Appendix K for more on the rapid evaluation of (z; ).

Exercises
1. Show that
d 01 (z)
= a(z) + b,
dz 1 (z)

(32.35)

(32.36)

a = 1,

b = e1 +

001 (1 /2)1 (1 /2) 01 (1 /2)2

,
1 (1 /2)2

where 1 = 1, 2 = .
2. In the setting of Exercise 1, deduce that a,b (z; ), given by (30.17), satisfies
01 (z a) 01 (z b)

1 (z a) 1 (z b)
d
1 (z a)
=
log
.
dz
1 (z b)

a,b (z; ) =
(32.37)

3. Show that, if a 6= ej ,
(32.38)

1
= A, (z) + B,
(z) a

where () = a. Show that

(32.39)

A=

1
0 ()

Identify B.
4. Give a similar treatment of 1/((z) a) for a = ej . Relate these functions to (z
j ),
with
j found from (32.19).
5. Express g2 and g3 , given in (31.17)(31.18), in terms of theta functions.
Hint. Use Exercise 7 of 31, plus Proposition 32.5

217

33. Elliptic integrals

The integral (31.8) is a special case of a class of integrals known as elliptic integrals,
which we explore in this section. Let us set
(33.1)

q() = ( e1 )( e2 )( e3 ).

We assume ej C are distinct and that (as in (31.15))

(33.2)

e1 + e2 + e3 = 0,

which could be arranged by a coordinate translation. Generally, an elliptic integral is an

integral of the form
Z

(33.3)

R(,

q()) d,

where R(, ) is a rational function of its arguments. The relevance of (31.8) is reinforced
by the following result.
Proposition 33.1. Given distinct ej satisfying (33.2), there exists a lattice , generated
by 1 , 2 C, linearly independent over R, such that if (z) = (z; ), then
(33.4)

= ej ,

1 j 3,

where 3 = 1 + 2 .
Given this result, we have from (31.7) that
0 (z)2 = 4q((z)),

(33.5)
and hence, as in (31.8),
(33.6)

1
2

(z)

(z0 )

d
p
= z z0 ,
q()

mod .

The problem of proving Proposition 33.1 is known as the Abel inversion problem. The
proof requires new tools, which will be provided in 34. We point out here that there is
no difficulty in identifying what the lattice must be. We have
(33.7)

1
2

ej

d
j
p
,
=
2
q()

mod ,

218

by (33.6). One can also verify directly from (33.7) that if the branches are chosen appropriately then 3 = 1 + 2 . It is not so clear that if is constructed directly from (33.7)
then the values of (z; ) at z = j /2 are given by (33.4), unless one already knows that
Proposition 33.1 is true.
Given Proposition 33.1, we can rewrite the elliptic integral (33.3) as follows. The result
depends on the particular path 01 from 0 to 1 and on the particular choice of path
01 in C/ such that maps 01 homeomorphically onto 01 . With these choices, (33.3)
becomes
Z

1
(33.8)
R (z), 0 (z) 0 (z) dz,
2
01

or, as we write more loosely,

Z

z1

(33.9)
z0

1
R (z), 0 (z) 0 (z) dz,
2

where z0 and z1 are the endpoints of 01 , satisfying (zj ) = j . It follows from Proposition
31.2 that

1 0 0
(33.10)
R (z), (z) (z) = R1 ((z)) + R2 ((z))0 (z),
2
for some rational functions Rj (). In fact, one can describe computational rules for producing such Rj , by using (33.5). Write R(, ) as a quotient of polynomials in (, ) and
use (33.5) to obtain that the left side of (33.10) is equal to
P1 ((z)) + P2 ((z))0 (z)
,
Q1 ((z)) + Q2 ((z))0 (z)

(33.11)

for some polynomials Pj (), Qj (). Then multiply the numerator and denominator of
(33.11) by Q1 ((z)) Q2 ((z))0 (z) and use (33.5) again on the new denominator to
obtain the right side of (33.10).
The integral of (33.3) is now transformed to the integral of the right side of (33.10).
Note that
Z z1
Z 1
0
(33.12)
R2 ((z)) (z) dz =
R2 () d, j = (zj ).
z0

Z z1
R1 ((z)) dz,
(33.13)
z0

219

We first analyze (33.13) when R1 () is a polynomial. To begin, we have

Z z1
(33.14)
(z) dz = (z0 ) (z1 ),
z0

by (30.15), where (z) = (z; ) is given by (30.14). See (32.35)(32.36) for a formula in
terms of theta functions. Next, differentiating (33.5) gives (as mentioned in Exercise 5 of
31)
1
00 (z) = 2q 0 ((z)) = 6(z)2 g2 ,
2

(33.15)
so
Z
(33.16)

z1

(z)2 dz = 0 (z1 ) 0 (z0 ) +

z0

g2
(z1 z0 ).
2

We can integrate (z)k+2 for k N via the following inductive procedure. We have
(33.17)

d 0
(z)(z)k = 00 (z)(z)k + k0 (z)2 (z)k1 .
dz

Apply (33.15) to 00 (z) and (33.5) (or equivalently (31.17)) to 0 (z)2 to obtain
d 0
(z)(z)k ) = (6 + 4k)(z)k+2 (3 + k)g2 (z)k kg3 (z)k1 .
dz
Rz
From here the inductive evaluation of z01 (z)k+2 dz, for k = 1, 2, 3, . . . , is straightforward.
To analyze (33.13) for a general rational function R1 (), we see upon making a partial
fraction decomposition that it remains to analyze
Z z1
(33.19)
((z) a)` dz,
(33.18)

z0

for ` = 1, 2, 3, . . . . One can also obtain inductive formulas here, by replacing (z)k by
((z) a)k in (33.18) and realizing that k need not be positive. We get
(33.20)

d 0
(z)((z) a)k = 00 (z)((z) a)k + k0 (z)2 ((z) a)k1 .
dz

Now write
(33.21)

0 (z)2 = 43 ((z) a)3 + 42 ((z) a)2 + 41 ((z) a) + 40 ,

00 (z) = 2A2 ((z) a)2 + 2A1 ((z) a) + 2A0 ,

where
(33.22)

j =

q (j) (a)
,
j!

Aj =

q (j+1) (a)
,
j!

220

to obtain

(33.23)

d 0
(z)((z) a)k
dz
= (2A2 + 4k3 )((z) a)k+2 + (2A1 + 4k2 )((z) a)k+1
+ (2A0 + 4k1 )((z) a)k + 4k0 ((z) a)k1 .

Note that
(33.24)

2A0 + 4k1 = (2 + 4k)q 0 (a).

0 = q(a),

Thus, if a is not equal to ej for any j and if we know the integral (33.19) for integral
` k (for some negative integer k), we can compute the integral for ` = 1 k, as long
as k 6= 0. If a = ej for some j, and if we know (33.19) for integral ` k 1, we can
compute it for ` = k, since then q 0 (a) 6= 0.
At this point, the remaining case of (33.19) to consider is the case ` = 1, i.e.,
Z

z1

(33.25)
z0

dz
.
(z) a

See Exercises 24 of 32 for expressions of ((z) a)1 in terms of logarithmic derivatives

of quotients of theta functions.
Note that the cases ` = 0, 1, and 1 of (33.19) are, under the correspondence of (33.3)
with (33.8), respectively equal to
Z

(33.26)
0

d
p
,
q()

1
0

d
( a) p
,
q()

1
d
p
.
a q()

These are called, respectively, elliptic integrals of the first, second, and third kind. The
material given above expresses the general elliptic integral (33.3) in terms of these cases.
There is another family of elliptic integrals, namely those of the form
Z
p
(33.27)
I = R(, Q()) d,
where R(, ) is a rational function of its arguments and Q() is a fourth degree polynomial:
(33.28)

Q() = ( a0 )( a1 )( a2 )( a3 ),

with aj C distinct. Such integrals can be transformed to integrals of the form (33.3),
via the change of variable
(33.29)

1
,
a0

d =

1
d.
2

221

One has

(33.30)

1
11
1
1

Q
+ a0 =
+ a0 a1
+ a0 a2
+ a0 a3

A
= 4 ( e1 )( e2 )( e3 ),

where
(33.31)

A = (a1 a0 )(a2 a0 )(a3 a0 ),

ej =

1
.
aj a0

Then we have
Z
(33.32)

I=

+ a0 ,

1
A p
q(
)
d,
2
2

with q( ) as in (33.1). After a further coordinate translation, one can alter ej to arrange
(33.2).
Elliptic integrals are frequently encountered in many areas of mathematics. Here we
give two examples, one from differential equations and one from geometry.
Our first example involves the differential equation for the motion of a simple pendulum,
which takes the form
(33.33)

d2
+ g sin = 0,
dt2

where ` is the length of the pendulum g the acceleration of gravity (32 ft./sec.2 on the
surface of the earth), and is the angle the pendulum makes with the downward-pointing
vertical axis. The total energy of the pendulum is proportional to
1 d 2 g
E=
cos .
2 dt
`

(33.34)

Applying d/dt to (33.34) and comparing with (33.33) shows that E is constant for each
solution to (33.33), so one has
r
1 d
g

(33.35)
= E + cos ,
`
2 dt
or
Z
(33.36)

d
= 2t + c,
E + a cos

Z
t
d
(33.37)
p
= + c0 ,
2
2
sin

222

with = E + a, = 2a. Then setting = sin , d = cos d, we have

Z
d
t
(33.38)
p
= + c0 ,
2
2
2
( )(1 )
which is an integral of the form (33.27). If instead in (33.36) we set = cos , so d =
sin d, we obtain
Z

d
= 2 t + c,
(33.39)
p
(E + a)(1 2 )
which is an integral of the form (33.3).
In our next example we produce a formula for the arc length L() of the portion of the
ellipse
(33.40)

from t = 0 to t = . We assume a > b > 0. Note that

|z 0 (t)|2 = a2 sin2 t + b2 cos2 t

(33.41)

= b2 + c2 sin2 t,

with c2 = a2 b2 , so
Z
(33.42)

L() =

p
b2 + c2 sin2 t dt.

With = sin t, u = sin , this becomes

Z up
d
b2 + c2 2 p
1 2
0
(33.43)
Z u
b2 + c2 2
p
=
d,
(1 2 )(b2 + c2 2 )
0
which is an integral of the form (33.27).

Exercises
1. Using (33.7) and the comments that follow it, show that, for j = 1, 2,
(33.44)

1
2

e3
ej

d
j 0
p
,
=
2
q()

mod ,

223

where j 0 = 2 if j = 1, j 0 = 1 if j = 2.
2. Setting ekj = ek ej , show that
(33.45)

1
2

e1 +

e1

X
d
1
1/2 1/2
1
k+`+1/2
p
=
2 e12 e13
k
`
ek12 e`13 k + ` + 1/2
q()
k,`=0

is a convergent power series provided || < min(|e1 e2 |, |e1 e3 |). Using this and variants
to integrate from ej to ej + for j = 2 and 3, find convergent power series for j /2 (mod
).
3. Given k 6= 1, show that
Z
Z
d
d
1
p
p
(33.46)
= p
,
(1 2 )(k 2 2 )
2(1 k 2 )
q( )
with
=

1
1
1
q( ) =

.
2
1k
1+k

1
,
+1

In Exercises 49, we assume ej are real and e1 < e2 < e3 . We consider

Z e2
d
p
3 =
.
( e1 )( e2 )( e3 )
e1
4. Show that
Z
(33.48)

/2

3 = 2

(e3 e2 ) sin2 + (e3 e1 ) cos2

= 2I e3 e2 , e3 e1 ,
0

where
Z
(33.49)

/2

I(r, s) =
0

d
2

r2 sin + s2 cos2

Note that in (33.48), 0 < r < s.

Exercises 57 will be devoted to showing that
(33.50)

I(r, s) =

,
2M (s, r)

224

(33.51)

(a0 , b0 ) = (a, b),

(ak+1 , bk+1 ) =

a + b p

k
k
, ak bk .
2

Show that
a0 a1 a2 b2 b1 b0 .
Show that

(33.52)

lim ak = lim bk = M (a, b),

the latter identity being the definition of M (a, b). Show also that
a2k+1 b2k+1 =

1
(ak bk )2 ,
4

ak+1 bk+1 =

(ak bk )2
.
8ak+2

hence
(33.53)

Deduce from (33.53) that convergence in (33.52) is quite rapid.

6. Show that the asserted identity (33.50) holds if it can be demonstrated that, for 0 <
r s,
(33.54)

I(r, s) = I

rs,

r + s
.
2

Hint. Show that (33.54) I(r, s) = I(m, m), with m = M (s, r), and evaluate I(m, m).
7. Take the following steps to prove (33.54). Show that you can make the change of
variable from to , with
(33.55)

sin =

2s sin
,
(s + r) + (s r) sin2

,
2

and obtain
Z
(33.56)

/2

I(r, s) =
0

2 d
q
.
4rs sin2 + (s + r)2 cos2

225

Show that this yields (33.54).

8. In the setting of Exercise 4, deduce that
(33.57)

3 =

.
M ( e3 e1 , e3 e2 )

9. Similarly, show that

Z

e3

1 =

( e1 )( e2 )( e3 )

e2

/2

(33.58)

(e2 e1 ) sin2 + (e3 e1 ) cos2

= 2i

.
M ( e3 e1 , e2 e1 )

Z /2

d
2

1 2 sin

dx

=
0

(1

x2 )(1

2 x2 )

Write 1 2 sin2 = (1 2 ) sin2 + cos2 to deduce that

Z 1

dx
p
p
=
,
(3.59)
(1 x2 )(1 2 x2 )
2M (1, 1 2 )
0
if (1, 1).
11. Parallel to Exercise 10, show that
Z /2
Z 1
d
dx
p
p
,
=
(1 x2 )(1 + 2 x2 )
0
0
1 + 2 sin2
and deduce that
Z

(3.60)
0

dx

p
p
=
,
(1 x2 )(1 + 2 x2 )
2M ( 1 + 2 , 1)

if R. A special case is
Z

(3.61)
0

dx

.
=
2M ( 2, 1)
1 x4

226

q()

Recall from 33 the cubic polynomial

(34.1)

q() = ( e1 )( e2 )( e3 ),

where e1 , e2 , e3 C are distinct. Here we will

pconstruct a compact Riemann surface M
associated with the double valued function q(), together with a holomorphic map
(34.2)

: M S 2 ,

and discuss some important properties of M and . We will then use this construction
to prove Proposition 33.1. Material developed below will use some basic results on manifolds, particularly on surfaces, which are generally covered in beginning topology courses.
Background may be found in [Mun] and [Sp], among other places. See also Appendix C
for some background.
To begin, we set e4 = in the Riemann sphere C {}, identified with S 2 in 26.
Reordering if necessary, we arrange that the geodesic 12 from e1 to e2 is disjoint from the
geodesic 34 from e3 to e4 . We slit S 2 along 12 and along 34 , obtaining X, a manifold
with boundary, as illustrated in the top right portion of Fig. 34.1. Now
(34.3)

M = X1 X2 / ,

where X1 and X2 are two copies of X, and the equivalence relation identifies the upper
boundary of X1 along the slit 12 with the lower boundary of X2 along this slit and viceversa, and similarly for 34 . This is illustrated in the middle and bottom parts of Fig. 34.1.
The manifold M is seen to be topologically equivalent to a torus.
The map : M S 2 in (34.2) is tautological. It is two-to-one except for the four
points pj = 1 (ej ). Recall the definition of a Riemann surface given in 26, in terms
of coordinate covers. The space M has a unique Riemann surface structure for which
is holomorphic. A coordinate taking a neighborhood of pj in M bijectively onto a
neighborhood of the origin in C is given by j (x) = ((x) ej )1/2 , for 1 j 3, with
(x) S 2 C {}, and a coordinate mapping a neighborhood of p4 in M bijectively
1/2
onto a neighborhood of the origin in C is given
.
pby 4 (x) =2 (x)
Now consider the double-valued form d/ q() on S , having singularities at {ej }.
This pulls back to a single-valued 1-form on M . Noting that if w2 = then
(34.4)

d
= 2 dw,

and that if w2 = 1/ then

(34.5)

d
p = 2 dw,
3

we see that is a smooth, holomorphic 1-form on M , with no singularities, and also that
has no zeros on M . Using this, we can prove the following.

227

(34.6)

: M C/0 .

Proof. Given M homeomorphic to S 1 S 1 , we have closed curves c1 and c2 through p1

in M such that each closed curve in M is homotopic to a curve Rstarting at p1 , winding
n1 times along c1 , then n2 times along c2 , with nj Z. Say j = cj . We claim 1 and
2 are linearly independent over R. First we show that they are not both 0. Indeed, if
1 = 2 = 0, then
Z z
(34.7)
(z) =

p0

would define a non-constant holomorphic map : M C, which would contradict

the
Rz
1
maximum principle. Let us say 2 6= 0, and set = 2 . Then 1 (z) = p0 is well
defined modulo an additive term of the form j + k(1 /2 ), with j, k Z. If 1 /2 were
real, then Im 1 : M R would be a well defined harmonic function, hence (by the
maximum principle) constant, forcing constant, and contradicting the fact that 6= 0.
Thus we have that 1 = {n1 1 + n2 2 : nj Z} is a lattice, and that (34.7) yields a
well defined holomorphic map
(34.8)

: M C/1 .

Since is nowhere vanishing, is a local diffeomorphism. Hence it must be a covering

map. This gives (34.6), where 0 is perhaps a sublattice of 1 .
We now prove Proposition 33.1, which we restate here.
Proposition 34.2. Let e1 , e2 , e3 be distinct points in C, satisfying
(34.9)

e1 + e2 + e3 = 0.

There exists a lattice C, generated by 1 , 2 , linearly independent over R, such that

if (z) = (z; ), then

j
(34.10)

= ej , 1 j 3,
2
where 3 = 1 + 2 .
Proof. We have from (34.2) and (34.6) a holomorphic map
(34.11)

: C/0 S 2 ,

which is a branched double cover, branching over e1 , e2 , e3 , and . We can regard as a

meromorphic function on C, satisfying
(34.12)

(z + ) = (z),

0 .

228

Furthermore, translating coordinates, we can assume has a double pole, precisely at

points in 0 . It follows that there are constants a and b such that
(34.13)

a C , b C,

(z) = a0 (z) + b,

where 0 (z) = (z; 0 ). Hence 0 (z) = a00 (z), so by Proposition 31.1 we have
(34.14)

0 (z) = 0 z =

0j
,
2

mod 0 ,

where 01 , 02 generate 0 and 03 = 01 + 02 . Hence (perhaps after some reordering)

(34.15)

ej = a0

0j

+ b.

Now if e0j = 0 (0j /2), we have by (31.15) that e01 + e02 + e03 = 0, so (34.9) yields
(34.16)

b = 0.

Finally, we set = a1/2 0 and use (32.34) to get

(34.17)

(z; ) = a (a1/2 z; 0 ).

Then (34.10) is achieved.

We mention that a similar construction works topyield a compact Riemann surface
M S 2 on which there is a single valued version of q() when
(34.18)

q() = ( e1 ) ( em ),

where ej C are distinct, and m 2. If m = 2g + 1, one has slits from e2j1 to e2j , for
j = 1, . . . , g, and a slit from e2g+1 to , which we denote e2g+2 . If m = 2g + 2, one has
slits from e2j1 to e2j , for j = 1, . . . , g + 1. Then X is constructed by opening the slits,
and M is constructed as in (34.3). The picture looks like that in Fig. 34.1, but instead of
two sets of pipes getting attached, one has g + 1 sets. One gets a Riemann
p surface M with
g holes, called a surface of genus g. Again the double-valued form d/ q() on S 2 pulls
back to a single-valued 1-form on M , with no singularities, except when m = 2 (see the
exercises). If m = 4 (so again g = 1), has no zeros. If m 5 (so g 2), has a zero
at 1 (). Proposition 34.1 extends to the case m = 4. If m 5 the situation changes.
It is a classical result that M is covered by the disk D rather than by C. The pull-back
of to D is called an automorphic form. For much more on such matters, and on more
general constructions of Riemann surfaces, we recommend [FK] and [MM].
We end this section with a brief description of a Riemann surface, conformally equivalent
to M in (34.3), appearing as a submanifold of complex projective space CP2 . More details
on such a construction can be found in [Cl] and [MM].

229

To begin, we define complex projective space CPn as (Cn+1 \ 0)/ , where we say z and
z 0 Cn+1 \ 0 satisfy z z 0 provided z 0 = az for some a C . Then CPn has the structure
of a complex manifold. Denote by [z] the equivalence class in CPn of z Cn+1 \ 0. We
note that the map
(34.19)

: CP1 C {}

given by
(34.20)

z2 6= 0,

([(1, 0)]) = ,

is a holomorphic diffeomorphism, so CP1 S 2 .

Now given distinct e1 , e2 , e3 C, we can define Me CP2 to consist of elements
[(w, , t)] such that (w, , t) C3 \ 0 satisfies
(34.21)

w2 t = ( e1 t)( e2 t)( e3 t).

One can show that Me is a smooth complex submanifold of CP2 , possessing then the
structure of a compact Riemann surface. An analogue of the map (34.2) is given as
follows.
Set p = [(1, 0, 0)] CP2 . Then there is a holomorphic map
(34.22)

: CP2 \ p CP1 ,

given by
(34.23)

This restricts to Me \ p CP1 . Note that p Me . While in (34.22) is actually singular

at p, for the restriction to Me \ p this is a removable singularity, and one has a holomorphic
map
(34.24)

e : Me CP1 C {} S 2 ,

given by (34.22) on Me \ p and taking p to [(1, 0)] CP1 , hence to C {}. This map
can be seen to be a 2-to-1 branched covering, branching over B = {e1 , e2 , e3 , }. Given
q C, q
/ B, and a choice r 1 (q) M and re 1
e (q) Me , there is a unique
holomorphic diffeomorphism
(34.25)
such that (r) = re and = e .

: M Me ,

230

Exercises
1. Show that the covering map in (34.8) is actually a diffeomorphism, and hence 0 = 1 .
2. Suppose 0 and 1 are two lattices in C such that T0 and T1 are conformally
equivalent, via a holomorphic diffeomorphism
(34.26)

f : C/0 C/1 .

Show that f lifts to a holomorphic diffeomorphism F of C onto itself, such that F (0) = 0,
and hence that F (z) = az for some a C . Deduce that 1 = a0 .
3. Consider the upper half-plane U = { C : Im > 0}. Given U , define
(34.27)

( ) = {m + n : m, n Z}.

Show that each lattice C has the form = a( ) for some a C , U.

4. Define the maps , : U U by
1
( ) = ,

(34.28)

( ) = + 1.

(34.29)

(( )) = 1 ( ),

(( )) = ( ).

5. Let G be the group of automorphisms of U generated by and , given in (34.28).

Show that if , 0 U ,
(34.30)

C/( ) C/( 0 ),

in the sense of being holomorphically diffeomorphic, if and only if

(34.31)

0 = ( ),

for some G.

6. Show that the group G consists of linear fractional transformations of the form

a + b
a b
(34.32)
LA ( ) =
, A=
,
c d
c + d
where a, b, c, d Z and det A = 1, i.e., A Sl(2, Z). Show that
G Sl(2, Z)/{I}.

231

In Exercises 78, we make use of the covering map : U C \ {0, 1}, given by (25.5),
and results of Exercises 18 of 25, including (25.10)(25.11), i.e.,
(34.33)

(( )) =

1
,
( )

(( )) = 1 ( ).

7. Given , 0 U, we say 0 if and only if (34.30) holds. Show that, given

(34.34)

, 0 U,

w = ( ), w0 = ( 0 ) C \ {0, 1},

we have
0 w0 = F (w) for some F G,

(34.35)

where G is the group (of order 6) of automorphisms of C \ {0, 1} arising in Exercise 6 of

25.
8. Bringing in the map H : S 2 S 2 arising in Exercise 8 of 25, i.e.,
(34.36)

H(w) =

4 (w2 w + 1)3
,
27 w2 (w 1)2

(34.37)

1
w

= H(w),

H(1 w) = H(w),

show that
(34.38)

Deduce that, for , 0 U ,

(34.39)

0 H ( 0 ) = H ( ).

Exercises 914 deal with the Riemann surface M of

(34.40)

q() when

q() = ( e1 )( e2 ),

and e1 , e2 C are distinct.

9. Show that the process analogous to that pictured in Fig. 34.1 involves the attachment
of one pair of pipes, and M is topologically equivalent to a sphere. One gets a branched
covering : M S 2 , as in (34.2).

232

p
10. Show that the double-valued form d/ q() on S 2 pulls back to a single-valued form
on M . Using (34.4), show that is a smooth nonvanishing form except at {p1 , p2 } =
1 (). In a local coordinate system about pj of the form j (x) = (x)1 , use a variant
of (34.4)(34.5) to show that has the form
= (1)j

(34.41)

g(z)
dz,
z

where g(z) is holomorphic and g(0) 6= 0.

11. Let c be a curve in M \ {p1 , p2 } with winding number 1 about p1 . Set
Z
(34.42)
= , L = {k : k Z} C.
c

Z z
(34.43)
(z) =

yields a well defined holomorphic map

(34.44)

: M \ {p1 , p2 } C/L.

12. Show that in (34.44) is a holomorphic diffeomorphism of M \ {p1 , p2 } onto C/L.

Hint. To show is onto, use (34.41) to examine the behavior of near p1 and p2 .
13. Produce a holomorphic diffeomorphism C/L C \ {0}, and then use (34.44) to obtain
a holomorphic diffeomorphism
1 : M \ {p1 , p2 } S 2 \ {0, }.

(34.45)

Show that this extends uniquely to a holomorphic diffeomorphism

1 : M S 2 .

(34.46)

14. Note that with a linear change of variable we can arrange ej = (1)j in (34.40). Relate
the results of Exercises 913 to the identity
Z z
(1 2 )1/2 d = sin1 z (mod 2Z).
(34.47)
0

233

A. Metric spaces, convergence, and compactness

A metric space is a set X, together with a distance function d : X X [0, ), having
the properties that
d(x, y) = 0 x = y,
(A.1)

d(x, y) = d(y, x),

d(x, y) d(x, z) + d(y, z).

The third of these properties is called the triangle inequality. An example of a metric space
is the set of rational numbers Q, with d(x, y) = |x y|. Another example is X = Rn , with
d(x, y) =

p
(x1 y1 )2 + + (xn yn )2 .

If (x ) is a sequence in X, indexed by = 1, 2, 3, . . . , i.e., by Z+ , one says x y if

d(x , y) 0, as . One says (x ) is a Cauchy sequence if d(x , x ) 0 as , .
One says X is a complete metric space if every Cauchy sequence converges to a limit in
X. Some metric spaces are not complete; for example,
Q is not complete. You can take a
sequence (x ) of rational numbers such that x 2, which is not rational. Then (x ) is
Cauchy in Q, but it has no limit in Q.
b as follows. Let
If a metric space X is not complete, one can construct its completion X
b consist of an equivalence class of Cauchy sequences in X, where we say
an element of X
(x ) (y ) provided d(x , y ) 0. We write the equivalence class containing (x ) as [x ].
If = [x ] and = [y ], we can set d(, ) = lim d(x , y ), and verify that this is well
b a complete metric space.
defined, and makes X
If the completion of Q is constructed by this process, you get R, the set of real numbers.
This construction provides a good way to develop the basic theory of the real numbers. A
detailed construction of R using this method is given in Chapter 1 of [T0].
There are a number of useful concepts related to the notion of closeness. We define
some of them here. First, if p is a point in a metric space X and r (0, ), the set
(A.2)

is called the open ball about p of radius r. Generally, a neighborhood of p X is a set

containing such a ball, for some r > 0.
A set U X is called open if it contains a neighborhood of each of its points. The
complement of an open set is said to be closed. The following result characterizes closed
sets.
Proposition A.1. A subset K X of a metric space X is closed if and only if
(A.3)

xj K, xj p X = p K.

234

Proof. Assume K is closed, xj K, xj p. If p

/ K, then p X \ K, which is open, so
some B (p) X \ K, and d(xj , p) for all j. This contradiction implies p K.
Conversely, assume (A.3) holds, and let q U = X \ K. If B1/n (q) is not contained in
U for any n, then there exists xn K B1/n (q), hence xn q, contradicting (A.3). This
completes the proof.
The following is straightforward.
Proposition A.2. If U is a family of open sets in X, then U is open. If K is a
family of closed subsets of X, then K is closed.
Given S X, we denote by S (the closure of S) the smallest closed subset of X
containing S, i.e., the intersection of all the closed sets K X containing S. The
following result is straightforward.
Proposition A.3. Given S X, p S if and only if there exist xj S such that xj p.
Given S X, p X, we say p is an accumulation point of S if and only if, for each
> 0, there exists q S B (p), q 6= p. It follows that p is an accumulation point of S if
and only if each B (p), > 0, contains infinitely many points of S. One straightforward
observation is that all points of S \ S are accumulation points of S.
The interior of a set S X is the largest open set contained in S, i.e., the union of all
the open sets contained in S. Note that the complement of the interior of S is equal to
the closure of X \ S.
We now turn to the notion of compactness. We say a metric space X is compact provided
the following property holds:
(A)

Each sequence (xk ) in X has a convergent subsequence.

We will establish various properties of compact metric spaces, and provide various equivalent characterizations. For example, it is easily seen that (A) is equivalent to:
(B)

The following property is known as total boundedness:

Proposition A.4. If X is a compact metric space, then
(C) Given > 0, finite set {x1 , . . . , xN } such that B (x1 ), . . . , B (xN ) covers X.
Proof. Take > 0 and pick x1 X. If B (x1 ) = X, we are done. If not, pick x2
X \ B (x1 ). If B (x1 ) B (x2 ) = X, we are done. If not, pick x3 X \ [B (x1 ) B (x2 )].
Continue, taking xk+1 X \ [B (x1 ) B (xk )], if B (x1 ) B (xk ) 6= X. Note
that, for 1 i, j k,
i 6= j = d(xi , xj ) .
If one never covers X this way, consider S = {xj : j N}. This is an infinite set with no
accumulation point, so property (B) is contradicted.

235

Corollary A.5. If X is a compact metric space, it has a countable dense subset.

Proof. Given = 2n , let Sn be a finite set of points xj such that {B (xj )} covers X.
Then C = n Sn is a countable dense subset of X.
Here is another useful property of compact metric spaces, which will eventually be
generalized even further, in (E) below.
Proposition A.6. Let X be a compact metric space. Assume K1 K2 K3 form
a decreasing sequence of closed subsets of X. If each Kn 6= , then n Kn 6= .
Proof. Pick xn Kn . If (A) holds, (xn ) has a convergent subsequence, xnk y. Since
{xnk : k `} Kn` , which is closed, we have y n Kn .
Corollary A.7. Let X be a compact metric space. Assume U1 U2 U3 form an
increasing sequence of open subsets of X. If n Un = X, then UN = X for some N .
Proof. Consider Kn = X \ Un .
The following is an important extension of Corollary A.7.
Proposition A.8. If X is a compact metric space, then it has the property:
(D)

Every open cover {U : A} of X has a finite subcover.

Proof. Each U is a union of open balls, so it suffices to show that (A) implies the following:
(D)

Let C = {zj : j N} X be a countable dense subset of X, as in Corollary A.2. Each B

is a union of balls Brj (zj ), with zj C B , rj rational. Thus it suffices to show that
(D)

For this, we set

Un = B1 Bn
and apply Corollary A.7.
The following is a convenient alternative to property (D):
(E)

Considering U = X \ K , we see that

(D) (E).
The following result completes Proposition A.8.

236

Theorem A.9. For a metric space X,

(A) (D).
Proof. By Proposition A.8, (A) (D). To prove the converse, it will suffice to show that
(E) (B). So let S X and assume S has no accumulation point. We claim:
Such S must be closed.
Indeed, if z S and z
/ S, then z would have to be an accumulation point. Say S = {x :
A}. Set K = S \ {x }. Then each K has no accumulation point, hence K X is
closed. Also K = . Hence there exists a finite set F A such that F K = , if
(E) holds. Hence S = F {x } is finite, so indeed (E) (B).
Remark. So far we have that for every metric space X,
(A) (B) (D) (E) = (C).
We claim that (C) implies the other conditions if X is complete. Of course, compactness
implies completeness, but (C) may hold for incomplete X, e.g., X = (0, 1) R.
Proposition A.10. If X is a complete metric space with property (C), then X is compact.
Proof. It suffices to show that (C) (B) if X is a complete metric space. So let S X
be an infinite set. Cover X by balls B1/2 (x1 ), . . . , B1/2 (xN ). One of these balls contains
infinitely many points of S, and so does its closure, say X1 = B1/2 (y1 ). Now cover X by
finitely many balls of radius 1/4; their intersection with X1 provides a cover of X1 . One
such set contains infinitely many points of S, and so does its closure X2 = B1/4 (y2 ) X1 .
Continue in this fashion, obtaining
X1 X2 X3 Xk Xk+1 ,

Xj B2j (yj ),

each containing infinitely many points of S. One sees that (yj ) forms a Cauchy sequence.
If X is complete, it has a limit, yj z, and z is seen to be an accumulation point of S.
If Xj , 1 j m, is a finite collection of metric spaces, with metrics dj , we can define
a Cartesian product metric space
(A.4)

X=

m
Y

Xj ,

d(x, y) = d1 (x1 , y1 ) + + dm (xm , ym ).

j=1

p
Another choice of metric is (x, y) = d1 (x1 , y1 )2 + + dm (xm , ym )2 . The metrics d and
are equivalent, i.e., there exist constants C0 , C1 (0, ) such that
(A.5)

x, y X.

A key example is Rm , the Cartesian product of m copies of the real line R.

We describe some important classes of compact spaces.

237

Qm
j=1

Xj .

Proof. If (x ) is an infinite sequence of points in X, say x = (x1 , . . . , xm ), pick a

convergent subsequence of (x1 ) in X1 , and consider the corresponding subsequence of
(x ), which we relabel (x ). Using this, pick a convergent subsequence of (x2 ) in X2 .
Continue. Having a subsequence such that xj yj in Xj for each j = 1, . . . , m, we then
have a convergent subsequence in X.
The following result is useful for calculus on Rn .
Proposition A.12. If K is a closed bounded subset of Rn , then K is compact.
Proof. The discussion above reduces the problem to showing that any closed interval I =
[a, b] in R is compact. This compactness is a corollary of Proposition A.10. For pedagogical
purposes, we redo the argument here, since in this concrete case it can be streamlined.
Suppose S is a subset of I with infinitely many elements. Divide I into 2 equal subintervals, I1 = [a, b1 ], I2 = [b1 , b], b1 = (a + b)/2. Then either I1 or I2 must contain infinitely
many elements of S. Say Ij does. Let x1 be any element of S lying in Ij . Now divide Ij in
two equal pieces, Ij = Ij1 Ij2 . One of these intervals (say Ijk ) contains infinitely many
points of S. Pick x2 Ijk to be one such point (different from x1 ). Then subdivide Ijk
into two equal subintervals, and continue. We get an infinite sequence of distinct points
x S, and |x x+k | 2 (b a), for k 1. Since R is complete, (x ) converges, say
to y I. Any neighborhood of y contains infinitely many points in S, so we are done.
If X and Y are metric spaces, a function f : X Y is said to be continuous provided
x x in X implies f (x ) f (x) in Y. An equivalent condition, which the reader is
invited to verify, is
(A.6)

U open in Y = f 1 (U ) open in X.

Proposition A.13. If X and Y are metric spaces, f : X Y continuous, and K X

compact, then f (K) is a compact subset of Y.
Proof. If (y ) is an infinite sequence of points in f (K), pick x K such that f (x ) = y .
If K is compact, we have a subsequence xj p in X, and then yj f (p) in Y.
If F : X R is continuous, we say f C(X). A useful corollary of Proposition A.13 is:
Proposition A.14. If X is a compact metric space and f C(X), then f assumes a
maximum and a minimum value on X.
Proof. We know from Proposition A.13 that f (X) is a compact subset of R. Hence f (X)
is bounded, say f (X) I = [a, b]. Repeatedly subdividing I into equal halves, as in the
proof of Proposition A.12, at each stage throwing out intervals that do not intersect f (X),
and keeping only the leftmost and rightmost interval amongst those remaining, we obtain
points f (X) and f (X) such that f (X) [, ]. Then = f (x0 ) for some x0 X
is the minimum and = f (x1 ) for some x1 X is the maximum.

238

If S R is a nonempty, bounded set, Proposition A.12 implies S is compact. The

function : S R, (x) = x is continuous, so by Proposition A.14 it assumes a maximum
and a minimum on S. We set
(A.7)

sup S = max x,

inf S = min x,

sS

xS

when S is bounded. More generally, if S R is nonempty and bounded from above, say
S (, B], we can pick A < B such that S [A, B] is nonempty, and set
(A.8)

sup S = sup S [A, B].

Similarly, if S R is nonempty and bounded from below, say S [A, ), we can pick
B > A such that S [A, B] is nonempty, and set
(A.9)

(A.10)

xX

(A.11)

inf f (x) = inf f (X).

xX

If f is not bounded from above, we set sup f = +, and if f is not bounded from below,
we set inf f = .
Given a set X, f : X R, and xn x, we set

(A.11A)
lim sup f (xn ) = lim sup f (xk ) ,
n

n kn

and
(A.11B)

n

inf f (xk ) .

n kn

We return to the notion of continuity. A function f C(X) is said to be uniformly

continuous provided that, for any > 0, there exists > 0 such that
(A.12)

An equivalent condition is that f have a modulus of continuity, i.e., a monotonic function

: [0, 1) [0, ) such that & 0 () & 0, and such that
(A.13)

x, y X, d(x, y) 1 = |f (x) f (y)| ().

Not all continuous functions are uniformly continuous. For example, if X = (0, 1) R,
then f (x) = sin 1/x is continuous, but not uniformly continuous, on X. The following
result is useful, for example, in the development of the Riemann integral.

239

Proposition A.15. If X is a compact metric space and f C(X), then f is uniformly

continuous.
Proof. If not, there exist x , y X and > 0 such that d(x , y ) 2 but
(A.14)

|f (x ) f (y )| .

Taking a convergent subsequence xj p, we also have yj p. Now continuity of f at

p implies f (xj ) f (p) and f (yj ) f (p), contradicting (A.14).
If X and Y are metric spaces, the space C(X, Y ) of continuous maps f : X Y has a
natural metric structure, under some additional hypotheses. We use

(A.15)
D(f, g) = sup d f (x), g(x) .
xX

This sup exists provided f (X) and g(X) are bounded subsets of Y, where to say B Y is
bounded is to say d : B B [0, ) has bounded image. In particular, this supremum
exists if X is compact. The following is a natural completeness result.
Proposition A.16. If X is a compact metric space and Y is a complete metric space,
then C(X, Y ), with the metric (A.9), is complete.
Proof. That D(f, g) satisfies the conditions to define a metric on C(X, Y ) is straightforward. We check completeness. Suppose (f ) is a Cauchy sequence in C(X, Y ), so, as
,

(A.16)
sup sup d f+k (x), f (x) 0.
k0 xX

Then in particular (f (x)) is a Cauchy sequence in Y for each x X, so it converges, say

to g(x) Y . It remains to show that g C(X, Y ) and that f g in the metric (A.9).
In fact, taking k in the estimate above, we have

(A.17)
sup d g(x), f (x) 0,
xX

i.e., f g uniformly. It remains only to show that g is continuous. For this, let xj x
in X and fix > 0. Pick N so that N < . Since fN is continuous, there exists J such
that j J d(fN (xj ), fN (x)) < . Hence

j J d g(xj ), g(x) d g(xj ), fN (xj ) + d fN (xj ), fN (x) + d fN (x), g(x) < 3.

This completes the proof.
In case Y = R, C(X, R) = C(X), introduced earlier in this appendix. The distance
function (A.15) can be written
D(f, g) = kf gksup ,

xX

240

kf ksup is a norm on C(X).

Generally, a norm on a vector space V is an assignment f 7 kf k [0, ), satisfying
kf k = 0 f = 0,

kaf k = |a| kf k,

kf + gk kf k + kgk,

given f, g V and a a scalar (in R or C). A vector space equipped with a norm is called a
normed vector space. It is then a metric space, with distance function D(f, g) = kf gk.
If the space is complete, one calls V a Banach space.
In particular, by Proposition A.16, C(X) is a Banach space, when X is a compact
metric space.
We next give a couple of slightly more sophisticated results on compactness. The following extension of Proposition A.11 is a special case of Tychonovs Theorem.
Q
Proposition A.17. If {Xj : j Z+ } are compact metric spaces, so is X = j=1 Xj .
Here, we can make X a metric space by setting
(A.18)

d(x, y) =

X
j=1

2j

dj (pj (x), pj (y))

,
1 + dj (pj (x), pj (y))

where pj : X Xj is the projection onto the jth factor. It is easy to verify that, if x X,
then x y in X, as , if and only if, for each j, pj (x ) pj (y) in Xj .
Proof. Following the argument in Proposition A.11, if (x ) is an infinite sequence of points
in X, we obtain a nested family of subsequences
(A.19)

(x ) (x1 ) (x2 ) (xj )

such that p` (xj ) converges in X` , for 1 ` j. The next step is a diagonal construction.
We set
(A.20)

= x X.

Then, for each j, after throwing away a finite number N (j) of elements, one obtains from
( ) a subsequence of the sequence (xj ) in (A.19), so p` ( ) converges in X` for all `.
Hence ( ) is a convergent subsequence of (x ).
The next result is a special case of Ascolis Theorem.
Proposition A.18. Let X and Y be compact metric spaces, and fix a modulus of continuity (). Then

(A.21)
C = f C(X, Y ) : d f (x), f (x0 ) d(x, x0 ) x, x0 X
is a compact subset of C(X, Y ).
Proof. Let (f ) be a sequence in C . Let be a countable dense subset of X, as in Corollary
A.5. For each x , (f (x)) is a sequence in Y, which hence has a convergent subsequence.

241

Using a diagonal construction similar to that in the proof of Proposition A.17, we obtain
a subsequence ( ) of (f ) with the property that (x) converges in Y, for each x ,
say
(A.22)

(x) (x),

for all x , where : Y.

So far, we have not used (A.21). This hypothesis will now be used to show that
converges uniformly on X. Pick > 0. Then pick > 0 such that () < /3. Since X is
compact, we can cover X by finitely many balls B (xj ), 1 j N, xj . Pick M so
large that (xj ) is within /3 of its limit for all M (when 1 j N ). Now, for any
x X, picking ` {1, . . . , N } such that d(x, x` ) , we have, for k 0, M,

d +k (x), (x) d +k (x), +k (x` ) + d +k (x` ), (x` )

(A.23)
+ d (x` ), (x)
/3 + /3 + /3.
Thus ( (x)) is Cauchy in Y for all x X, hence convergent. Call the limit (x), so we
now have (A.22) for all x X. Letting k in (A.23) we have uniform convergence of
to . Finally, passing to the limit in
(A.24)

d( (x), (x0 )) (d(x, x0 ))

gives C .
We want to re-state Proposition A.18, bringing in the notion of equicontinuity. Given
metric spaces X and Y , and a set of maps F C(X, Y ), we say F is equicontinuous at a
point x0 X provided
(A.25)

> 0, > 0 such that x X, f F,

dX (x, x0 ) < = dY (f (x), f (x0 )) < .

We say F is equicontinuous on X if it is equicontinuous at each point of X. We say F is

uniformly equicontinuous on X provided
(A.26)

> 0, > 0 such that x, x0 X, f F,

dX (x, x0 ) < = dY (f (x), f (x0 )) < .

Note that (A.26) is equivalent to the existence of a modulus of continuity such that
F C , given by (A.21). It is useful to record the following result.
Proposition A.19. Let X and Y be metric spaces, F C(X, Y ). Assume X is compact.
then
(A.27)

F equicontinuous = F is uniformly equicontinuous.

242

Proof. The argument is a variant of the proof of Proposition A.15. In more detail, suppose
there exist x , x0 X, > 0, and f F such that d(x , x0 ) 2 but
(A.28)

d(f (x ), f (x0 )) .

Taking a convergent subsequence xj p X, we also have x0j p. Now equicontinuity

of F at p implies that there esists N < such that
(A.29)

d(g(xj ), g(p)) <

,
2

j N, g F,

Putting together Propositions A.18 and A.19 then gives the following.
Proposition A.20. Let X and Y be compact metric spaces. If F C(X, Y ) is equicontinuous on X, then it has compact closure in C(X, Y ).
We next define the notion of a connected space. A metric space X is said to be connected
provided that it cannot be written as the union of two disjoint nonempty open subsets.
The following is a basic class of examples.
Proposition A.21. Each interval I in R is connected.
Proof. Suppose A I is nonempty, with nonempty complement B I, and both sets are
open. Take a A, b B; we can assume a < b. Let = sup{x [a, b] : x A} This
exists, as a consequence of the basic fact that R is complete.
Now we obtain a contradiction, as follows. Since A is closed A. But then, since A
is open, there must be a neighborhood ( , + ) contained in A; this is not possible.
We say X is path-connected if, given any p, q X, there is a continuous map : [0, 1]
X such that (0) = p and (1) = q. It is an easy consequence of Proposition A.21 that X
is connected whenever it is path-connected.
The next result, known as the Intermediate Value Theorem, is frequently useful.
Proposition A.22. Let X be a connected metric space and f : X R continuous.
Assume p, q X, and f (p) = a < f (q) = b. Then, given any c (a, b), there exists z X
such that f (z) = c.
Proof. Under the hypotheses, A = {x X : f (x) < c} is open and contains p, while
B = {x X : f (x) > c} is open and contains q. If X is connected, then A B cannot be
all of X; so any point in its complement has the desired property.
The next result is known as the Contraction Mapping Principle, and it has many uses
in analysis. In particular, we will use it in the proof of the Inverse Function Theorem, in
Appendix B.

243

(A.30)

d(T x, T y) r d(x, y),

for some r < 1. (We say T is a contraction.) Then T has a unique fixed point x. For any
y0 X, T k y0 x as k .
Proof. Pick y0 X and let yk = T k y0 . Then d(yk , yk+1 ) rk d(y0 , y1 ), so
d(yk , yk+m ) d(yk , yk+1 ) + + d(yk+m1 , yk+m )
(A.31)

(rk + + rk+m1 ) d(y0 , y1 )

rk (1 r)1 d(y0 , y1 ).

It follows that (yk ) is a Cauchy sequence, so it converges; yk x. Since T yk = yk+1 and

T is continuous, it follows that T x = x, i.e., x is a fixed point. Uniqueness of the fixed
point is clear from the estimate d(T x.T x0 ) r d(x, x0 ), which implies d(x, x0 ) = 0 if x and
x0 are fixed points. This proves Theorem A.23.

Exercises
1. If X is a metric space, with distance function d, show that
|d(x, y) d(x0 , y 0 )| d(x, x0 ) + d(y, y 0 ),
and hence
d : X X [0, ) is continuous.
2. Let : [0, ) [0, ) be a C 2 function. Assume
(0) = 0,

0 > 0,

00 < 0.

Prove that if d(x, y) is symmetric and satisfies the triangle inequality, so does
(x, y) = (d(x, y)).
Hint. Show that such satisfies (s + t) (s) + (t), for s, t R+ .
3. Show that the function d(x, y) defined by (A.18) satisfies (A.1).
Hint. Consider (r) = r/(1 + r).
4. Let X be a compact metric space. Assume fj , f C(X) and
fj (x) % f (x),

x X.

244

Prove that fj f uniformly on X. (This result is called Dinis theorem.)

Hint. For > 0, let Kj () = {x X : f (x)fj (x) }. Note that Kj () Kj+1 () .
5. In the setting of (A.4), let
n
o1/2
(x, y) = d1 (x1 , y1 )2 + + dm (xm , ym )2
.
Show that
(x, y) d(x, y)

m (x, y).

6. Let X and Y be compact metric spaces. Show that if F C(X, Y ) is compact, then F
is equicontinuous. (This is a converse to Proposition A.20.)
7. Recall that a Banach space is a complete normed linear space. Consider C 1 (I), where
I = [0, 1], with norm
kf kC 1 = sup |f | + sup |f 0 |.
I

Show that C 1 (I) is a Banach space.

8. Let F = {f C 1 (I) : kf kC 1 1}. Show that F has compact closure in C(I). Find a
function in the closure of F that is not in C 1 (I).

245

B. Derivatives and diffeomorphisms

To start this section off, we define the derivative and discuss some of its basic properties.
Let O be an open subset of Rn , and F : O Rm a continuous function. We say F is
differentiable at a point x O, with derivative L, if L : Rn Rm is a linear transformation
such that, for y Rn , small,
(B.1)

F (x + y) = F (x) + Ly + R(x, y)

with
(B.2)

kR(x, y)k
0 as y 0.
kyk

We denote the derivative at x by DF (x) = L. With respect to the standard bases of Rn

and Rm , DF (x) is simply the matrix of partial derivatives,

Fj
(B.3)
DF (x) =
,
xk
so that, if v = (v1 , . . . , vn )t , (regarded as a column vector) then
(B.4)

DF (x)v =

X F
k

xk

vk , . . . ,

X Fm
k

xk

vk

It will be shown below that F is differentiable whenever all the partial derivatives exist
and are continuous on O. In such a case we say F is a C 1 function on O. More generally,
F is said to be C k if all its partial derivatives of order k exist and are continuous. If F
is C k for all k, we say F is C .
In (B.2), we can use the Euclidean norm on Rn and Rm . This norm is defined by
(B.5)

1/2
kxk = x21 + + x2n

for x = (x1 , . . . , xn ) Rn . Any other norm would do equally well.

We now derive the chain rule for the derivative. Let F : O Rm be differentiable at
x O, as above, let U be a neighborhood of z = F (x) in Rm , and let G : U Rk be
differentiable at z. Consider H = G F. We have

(B.6)

= G(z) + DG(z) DF (x)y + R(x, y) + R1 (x, y)

= G(z) + DG(z)DF (x)y + R2 (x, y)

246

with

0 as y 0.
kyk

(B.7)

Another useful remark is that, by the Fundamental Theorem of Calculus, applied to

(t) = F (x + ty),
Z
(B.8)

F (x + y) = F (x) +

DF (x + ty)y dt,
0

provided F is C 1 . A closely related application of the Fundamental Theorem of Calculus

is that, if we assume F : O Rm is differentiable in each variable separately, and that
each F/xj is continuous on O, then
n
n
X
X

F (x + y) = F (x) +
F (x + zj ) F (x + zj1 ) = F (x) +
Aj (x, y)yj ,

(B.9)

j=1

j=1

Aj (x, y) =
0

F
x + zj1 + tyj ej dt,
xj

where z0 = 0, zj = (y1 , . . . , yj , 0, . . . , 0), and {ej } is the standard basis of Rn . Now (B.9)
implies F is differentiable on O, as we stated below (B.4). Thus we have established the
following.
Proposition B.1. If O is an open subset of Rn and F : O Rm is of class C 1 , then F
is differentiable at each point x O.
As is shown in many calculus texts, one can use the Mean Value Theorem instead of
the Fundamental Theorem of Calculus, and obtain a slightly sharper result.
For the study of higher order derivatives of a function, the following result is fundamental.
Proposition B.2. Assume F : O Rm is of class C 2 , with O open in Rn . Then, for
each x O, 1 j, k n,
(B.10)

F
F
(x) =
(x).
xj xk
xk xj

To prove Proposition B.2, it suffices to treat real valued functions, so consider f : O R.

For 1 j n, set
(B.11)

j,h f (x) =

1
f (x + hej ) f (x) ,
h

247

where {e1 , . . . , en } is the standard basis of Rn . The mean value theorem (for functions of
xj alone) implies that if j f = f /xj exists on O, then, for x O, h > 0 sufficiently
small,
(B.12)

j,h f (x) = j f (x + j hej ),

for some j (0, 1), depending on x and h. Iterating this, if j (k f ) exists on O, then,
for x O and h > 0 sufficiently small,
k,h j,h f (x) = k (j,h f )(x + k hek )
(B.13)

= j,h (k f )(x + k hek )

= j k f (x + k hek + j hej ),

with j , k (0, 1). Here we have used the elementary result

(B.14)

k j,h f = j,h (k f ).

We deduce the following.

Proposition B.3. If k f and j k f exist on O and j k f is continuous at x0 O, then
(B.15)

h0

Clearly
(B.16)

so we have the following, which easily implies Proposition B.2.

Corollary B.4. In the setting of Proposition B.3, if also j f and k j f exist on O and
k j f is continuous at x0 , then
(B.17)

j k f (x0 ) = k j f (x0 ).

If U and V be open subsets of Rn and F : U V is a C 1 map, we say F is a

diffeomorphism of U onto V provided F maps U one-to-one and onto V , and its inverse
G = F 1 is a C 1 map. If F is a diffeomorphism, it follows from the chain rule that DF (x)
is invertible for each x U . We now present a partial converse of this, the Inverse Function
Theorem, which is a fundamental result in multivariable calculus.
Theorem B.5. Let F be a C k map from an open neighborhood of p0 Rn to Rn , with
q0 = F (p0 ). Assume k 1. Suppose the derivative DF (p0 ) is invertible. Then there is a
neighborhood U of p0 and a neighborhood V of q0 such that F : U V is one-to-one and
onto, and F 1 : V U is a C k map. (So F : U V is a diffeomorphism.)
First we show that F is one-to-one on a neighborhood of p0 , under these hypotheses.
In fact, we establish the following result, of interest in its own right.

248

Proposition B.6. Assume Rn is open and convex, and let f : Rn be C 1 .

Assume that the symmetric part of Df (u) is positive-definite, for each u . Then f is
one-to-one on .
Proof. Take distinct points u1 , u2 , and set u2 u1 = w. Consider : [0, 1] R,
given by
(t) = w f (u1 + tw).
Then 0 (t) = w Df (u1 + tw)w > 0 for t [0, 1], so (0) 6= (1). But (0) = w f (u1 )
and (1) = w f (u2 ), so f (u1 ) 6= f (u2 ).
To continue the proof of Theorem B.5, let us set

(B.18)
f (u) = A F (p0 + u) q0 , A = DF (p0 )1 .
Then f (0) = 0 and Df (0) = I, the identity matrix. We show that f maps a neighborhood
of 0 one-to-one and onto some neighborhood of 0. Proposition B.4 applies, so we know f
is one-to-one on some neighborhood O of 0. We next show that the image of O under f
contains a neighborhood of 0.
Note that
(B.19)

f (u) = u + R(u),

R(0) = 0, DR(0) = 0.

(B.20)

f (u) = v.

(B.21)

Tv (u) = v R(u).

Thus solving (B.20) is equivalent to solving

(B.22)

Tv (u) = u.

We look for a fixed point u = K(v) = f 1 (v). Also, we want to prove that DK(0) = I, i.e.,
that K(v) = v + r(v) with r(v) = o(kvk), i.e., r(v)/kvk 0 as v 0. If we succeed in
doing this, it follows easily that, for general x close to q0 , G(x) = F 1 (x) is defined, and

(B.23)

1
DG(x) = DF G(x)
.

Then a simple inductive argument shows that G is C k if F is C k .

A tool we will use to solve (B.22) is the Contraction Mapping Principle, established in
Appendix A, which states that if X is a complete metric space, and if T : X X satisfies
(B.24)

dist(T x, T y) r dist(x, y),

249

for some r < 1 (we say T is a contraction), then T has a unique fixed point x.
In order to implement this, we consider
(B.25)

Tv : Xv Xv

with
(B.26)

Xv = {u : ku vk Av }

where we set
(B.27)

Av =

sup

kR(w)k.

kwk2kvk

We claim that (B.25) holds if kvk is sufficiently small. To prove this, note that Tv (u) v =
R(u), so we need to show that, provided kvk is small, u Xv implies kR(u)k Av . But
indeed, if u Xv , then kuk kvk + Av , which is 2kvk if kvk is small, so then
kR(u)k

sup

kR(w)k = Av .

kwk2kvk

This establishes (B.25).

Note that Tv (u1 ) Tv (u2 ) = R(u2 ) R(u1 ), and R is a C k map, satisfying DR(0) = 0.
It follows that, if kvk is small enough, the map (B.18) is a contraction map. Hence there
exists a unique fixed point u = K(v) Xv . Also, since u Xv ,
(B.28)

kK(v) vk Av = o(kvk),

so the Inverse Function Theorem is proved.

Thus if DF is invertible on the domain of F, F is a local diffeomorphism. Stronger
hypotheses are needed to guarantee that F is a global diffeomorphism onto its range.
Proposition B.6 provides one tool for doing this. Here is a slight strengthening.
Corollary B.7. Assume Rn is open and convex, and that F : Rn is C 1 . Assume
there exist n n matrices A and B such that the symmetric part of A DF (u) B is positive
definite for each u . Then F maps diffeomorphically onto its image, an open set in
Rn .
Proof. Exercise.

250

C. Surfaces and metric tensors

A smooth m-dimensional surface M Rn is characterized by the following property.
Given p M, there is a neighborhood U of p in M and a smooth map : O U, from
an open set O Rm bijectively to U, with injective derivative at each point. Such a map
is called a coordinate chart on M. We call U M a coordinate patch. If all such maps
are smooth of class C k , we say M is a surface of class C k .
There is an abstraction of the notion of a surface, namely the notion of a manifold,
which we briefly mention at the end of this appendix.
If : O U is a C k coordinate chart, such as described above, and (x0 ) = p, we set
(C.1)

Tp M = Range D(x0 ),

a linear subspace of Rn of dimension m, and we denote by Np M its orthogonal complement.

It is useful to consider the following map. Pick a linear isomorphism A : Rnm Np M ,
and define
(C.2)

: O Rnm Rn ,

Thus is a C k map defined on an open subset of Rn . Note that

v
(C.3)
D(x0 , 0)
= D(x0 )v + Aw,
w
so D(x0 , 0) : Rn Rn is surjective, hence bijective, so the Inverse Function Theorem
applies; maps some neighborhood of (x0 , 0) diffeomorphically onto a neighborhood of
p Rn .
Suppose there is another C k coordinate chart, : U . Since and are by
hypothesis one-to-one and onto, it follows that F = 1 : O is a well defined
map, which is one-to-one and onto. See Fig. C.1. In fact, we can say more.
Lemma C.1. Under the hypotheses above, F is a C k diffeomorphism.
Proof. It suffices to show that F and F 1 are C k on a neighborhood of x0 and y0 , respectively, where (x0 ) = (y0 ) = p. Let us define a map in a fashion similar to (5.2).
ep M be its orthogonal complement.
To be precise, we set Tep M = Range D(y0 ), and let N
(Shortly we will show that Tep M = Tp M , but we are not quite ready for that.) Then pick a
ep M and set (y, z) = (y) + Bz, for (y, z) Rnm .
linear isomorphism B : Rnm N
Again, is a C k diffeomorphism from a neighborhood of (y0 , 0) onto a neighborhood of p.
It follows that 1 is a C k diffeomeophism from a neighborhood of (x0 , 0) onto a
neighborhood of (y0 , 0). Now note that, for x close to x0 and y close to y0 ,

(C.4)
1 (x, 0) = F (x), 0 , 1 (y, 0) = F 1 (y), 0 .

251

These identities imply that F and F 1 have the desired regularity.

Thus, when there are two such coordinate charts, : O U, : U , we have a
C diffeomorphism F : O such that
k

(C.5)

= F.

(C.6)

D(x) = D(y) DF (x),

y = F (x).

In particular this implies that Range D(x0 ) = Range D(y0 ), so Tp M in (C.1) is independent of the choice of coordinate chart. It is called the tangent space to M at p.
We next define an object called the metric tensor on M . Given a coordinate chart
: O U , there is associated an m m matrix G(x) = gjk (x) of functions on O,
defined in terms of the inner product of vectors tangent to M :
n

X ` `

gjk (x) = D(x)ej D(x)ek =

=
,
xj xk
xj xk

(C.7)

`=1

where {ej : 1 j m} is the standard orthonormal basis of Rm . Equivalently,

G(x) = D(x)t D(x).

(C.8)

We call (gjk ) the metric tensor of M, on U , with respect to the coordinate chart : O U .
Note that this matrix is positive-definite. From a coordinate-independent point of view,
the metric tensor on M specifies inner products of vectors tangent to M , using the inner
product of Rn .
If we take another coordinate chart : U, we want to compare (gjk ) with H =
(hjk ), given by
(C.9)

As seen above we have a diffeomorphism F : O such that (5.5)(5.6) hold. Consequently,

G(x) = DF (x)t H(y) DF (x),

(C.10)
or equivalently,
(C.11)

gjk (x) =

X Fi F`
hi` (y).
xj xk
i,`

We now define the notion of surface integral on M . If f : M R is a continuous

function supported on U, we set
Z
Z
p
(C.12)
f dS = f (x) g(x) dx,
M

252

where
(C.13)

We need to know that this is independent of the choice of coordinate chart

pU. Thus,
R :O
if we use : U instead, we want to show that (C.12) is equal to f (y) h(y) dy,
where h(y) = det H(y). Indeed, since f F = f , we can apply the change of variable
formula for multidimensional integrals, to get
Z
Z
p
p
(C.14)
f (y) h(y) dy = f (x) h(F (x)) |det DF (x)| dx.

Now, (C.10) implies that

p
p
g(x) = |det DF (x)| h(y),

(C.15)

so the right side of (C.14) is seen to be equal to (C.12), and our surface integral is well
defined, at least for f supported in a coordinate patch. More generally, if f : M R has
compact support, write it as a finite sum of terms, each supported on a coordinate patch,
and use (C.12) on each patch.
Let us pause to consider the special cases m = 1 and m = 2. For m = 1, we are
considering a curve in Rn , say : [a, b] Rn . Then G(x) is a 1 1 matrix, namely
G(x) = |0 (x)|2 . If we denote the curve in Rn by , rather than M, the formula (C.12)
becomes
Z
Z b
(C.16)
f ds =
f (x) |0 (x)| dx.
a

In case m = 2, let us consider a surface M R3 , with a coordinate chart : O U M.

For f supported in U, an alternative way to write the surface integral is
Z
Z
(C.17)
f dS = f (x) |1 2 | dx1 dx2 ,
M

where u v is the cross product

integrand with the one in (C.12).

1 1
(C.18)
g = det
2 1

of vectors u and v in R3 . To see this, we compare this

In this case,

1 2
= |1 |2 |2 |2 (1 2 )2 .
2 2

Recall that |u v| = |u| |v| | sin |, where is the angle between u and v. Equivalently,
since u v = |u| |v| cos ,

(C.19)
|u v|2 = |u|2 |v|2 1 cos2 = |u|2 |v|2 (u v)2 .

253

Thus we see that |1 2 | = g, in this case, and (C.17) is equivalent to (C.12).

1
An important class of surfaces is the class of graphs of smooth functions. Let u
C (),
n1
for an open R
, and let M be the graph of z = u(x). The map (x) = x, u(u)
provides a natural coordinate system, in which the metric tensor is given by
(C.20)

gjk (x) = jk +

u u
.
xj xk

If u is C 1 , we see that gjk is continuous. To calculate g = det(gjk ), at a given point p ,

if u(p) 6= 0, rotate coordinates so that u(p) is parallel to the x1 axis. We see that

(C.21)

1/2
g = 1 + |u|2
.

Z
Z

1/2
(C.22)
Vn1 (M ) = dS =
1 + |u(x)|2
dx.
M

Particularly important examples of surfaces are the unit spheres S n1 in Rn ,

S n1 = {x Rn : |x| = 1}.

(C.23)

Spherical polar coordinates on Rn are defined in terms of a smooth diffeomorphism

(C.24)

R : (0, ) S n1 Rn \ 0,

R(r, ) = r.

Let (h`m ) denote the metric tensor on S n1 (induced from its inclusion in Rn ) with respect
to some coordinate chart : O U S n1 . Then, with respect to the coordinate chart
: (0, ) O U Rn given by (r, y) = r(y), the Euclidean metric tensor can be
written

1
(C.25)
ejk =
.
r2 h`m
To see that the blank terms vanish, i.e., r xj = 0, note that (x) (x) = 1
xj (x) (x) = 0. Now (C.25) yields

(C.26)

e = rn1 h.

We therefore have the following result for integrating a function in spherical polar coordinates.
Z
Z hZ
i
(C.27)
f (x) dx =
f (r)rn1 dr dS().
Rn

S n1

254

We next compute the (n 1)-dimensional area An1 of the unit sphere S n1 Rn ,

using (C.27) together with the computation
Z
2

e|x| dx = n/2 ,

(C.28)
Rn

which can be reduced to the case n = 2 and done there in polar coordinates. First note
that, whenever f (x) = (|x|), (C.27) yields
Z

Z
(|x|) dx = An1

(C.29)

(r)rn1 dr.

Rn
2

In particular, taking (r) = er and using (C.28), we have

Z
(C.30)

n/2

= An1

r 2 n1

1
dr = An1
2

es sn/21 ds,

where we used the substitution s = r2 to get the last identity. We hence have
(C.31)

An1 =

2 n/2
,
(n/2)

Z
(C.32)

(z) =

es sz1 ds.

The gamma function is developed in 18.

Having discussed surfaces in Rn , we turn to the more general concept of a manifold,
useful for the construction in 34. A manifold is a metric space with an atlas, i.e., a
covering by open sets Uj together with homeomorphisms j : Uj Vj , Vj open in Rn .
The number n is called the dimension of M. We say that M is a smooth manifold of class
C k provided the atlas has the following property. If Ujk = Uj Uk 6= , then the map
jk : j (Ujk ) k (Ujk )
k
given by k 1
j , is a smooth diffeomorphism of class C from the open set j (Ujk ) to
the open set k (Ujk ) in Rn . By this, we mean that jk is C k , with a C k inverse. The pairs
(Uj , j ) are called local coordinate charts.
A continuous map from M to another smooth manifold N is said to be smooth of class
k
C if it is smooth of class C k in local coordinates. Two different atlasses on M, giving a
priori two structures of M as a smooth manifold, are said to be equivalent if the identity
map on M is smooth (of class C k ) from each one of these two manifolds to the other. Really

255

a smooth manifold is considered to be defined by equivalence classes of such atlasses, under

this equivalence relation.
It follows from Lemma C.1 that a C k surface in Rn is a smooth manifold of class C k .
Other examples of smooth manifolds include tori T/, introduced in 26 and used in 34, as
well as other Riemann surfaces discussed in 34. One can find more material on manifolds
in [Sp] and in [T2].
The notion of a metric tensor generalizes readily from surfaces in Rn to smooth manifolds; this leads to the notion of an integral on a manifold with a metric tensor (i.e., a
Riemannian manifold). For use in Appendix D, we give details about metric tensors, in
case M is covered by one coordinate chart,
(C.33)

1 : U1 M,

with U1 Rn open. In such a case, a metric tensor on M is defined by an n n matrix

(C.34)
G(x) = gjk (x) , gjk C k (U1 ),
which is taken to be symmetric and positive definite, generalizing the set-up in (C.7). If
there is another covering of M by a coordinate chart 2 : U2 M , a positive definite
matrix H on U2 defines the same metric tensor on M provided
G(x) = DF (x)t H(y)DF (x),

(C.35)

for y = F (x),

as in (C.10), where F is the diffeomorphism

F = 1
2 1 : U1 U2 .

(C.36)

We also say that G defines a metric tensor on U1 and H defines a metric tensor on U2 ,
and the diffeomorphism F : U1 U2 pulls H back to G.
Let : [a, b] U1 be a C 1 curve. The following is a natural generalization of (C.16),
defining the integral of a function f C(U1 ) over , with respect to arc length:
Z
Z b
h
i1/2
0
0
(C.37)
f ds =
f ((t)) (t) G((t)) (t)
dt.
a

If = F is the associated curve on U2 and if f = f F C(U2 ), we have

Z
Z b
h
i1/2

f ds =
f(
(t)) 0 (t) H(
(t))
0 (t)
dt
a

h
i1/2
f ((t)) DF ((t)) 0 (t) H(
(t))DF ((t)) 0 (t)
dt

h
i1/2
f ((t)) 0 (t) G(x) 0 (t)
dt

=
(C.38)

=
Za
=

f ds,

256

the second identity by the chain rule 0 (t) = DF ((t)) 0 (t) and the third identity by (C.35).
Another property of this integral is parametrization invariance. Say : [, ] [a, b] is
an order preserving C 1 diffeomorphism and = : [, ] U1 . Then
Z

f ds =

h
i1/2
f ((t)) 0 (t) G((t)) 0 (t)
dt

(C.39)

Z
=

h
i1/2
dt
f ( (t)) 0 (t)2 0 ((t)) G( (t)) 0 ((t))
h
i1/2
0
0
f (( )) ( ) G(( )) ( )
d

Za
=

f ds,

the second identity by the chain rule 0 (t) = 0 (t) 0 ((t)) and the third identity via the
change of variable = (t), d = 0 (t) dt.
The arc length of these curves is defined by integrating 1. We have
Z bh
(C.40)

`G () =

i1/2
0 (t) G((t)) 0 (t)
dt,

and a parallel definition of `H (

). With , , and related as in (C.38)(C.39), we have
(C.41)

`G () = `G () = `H (
).

Another useful piece of notation associated with the metric tensor G is

X
gjk (x) dxj dxk .
(C.42)
ds2 =
j,k

(C.43)

In case
(C.44)

we have
(C.45)

(C.46)

257

or
(C.47)

ds = A(z) |dz|.

Z
(C.48)

`G () =

A((t)) | 0 (t)| dt.

Under the change of variable y = F (x), the formula (C.35) for the metric tensor H =
(hjk ) on U2 that pulls back to G under F : U1 U2 is equivalent to
(C.49)

X
j,k

X
j,k

gjk (x) dxj dxk ,

dyj =

X Fj
`

x`

dx` .

258

D. Greens theorem
Here we prove Greens theorem, which was used in one approach to the Cauchy integral
theorem in 5.
Theorem D.1. If is a bounded region in R2 with piecewise smooth boundary, and
f, g C 1 (), then
ZZ
Z
g
f

dx dy = (f dx + g dy).
x y

(D.1)

The identity (D.1) combines two separate identities, namely

Z

ZZ

(D.2)

f
dx dy,
y

f dx =

and
Z
(D.3)

ZZ
g dy =

g
dx dy.
x

We will first prove (D.2) for domains of the form depicted in Fig. D.1 (which we will
call type I), then prove (D.3) for domains of the form depicted in Fig. D.2 (which we call
type II), and then discuss more general domains.
If is type I (cf. Fig. D.1), then
Z
(D.4)

f dx =
a

f (x, 0 (x)) dx

a

Z
(D.5)

1 (x)

0 (x)

f
(x, y) dy,
y

Z
(D.6)

1 (x)

0 (x)

f
(x, y) dy dx =
y

ZZ

f
dx dy,
y

259

Z

Z
g dy =

g(1 (y)) dy
c

(D.7)

g(0 (y), y) dy
c

1 (y)

=
c

0 (y)

ZZ

g
(x, y), dx dy
x

g
dx dy,
x

and we have (D.3).

Figure D.3 depicts a region that is type I but not type II. The argument above gives
(D.2) in this case, but we need a further argument to get (D.3). As indicated in the figure,
we divide into two pieces, 1 and 2 , and observe that 1 and 2 are each of type II.
Hence, given g C 1 (),
Z
ZZ
g
(D.8)
dx dy,
g dy =
y
j

for each j. Now

(D.9)

ZZ
X Z Z g
g
dx dy =
dx dy.
y
y
j
j

R
On the other hand, if we sum the integrals j g dy, we get an integral over plus two
integrals over the interface between 1 and 2 . However, the latter two integrals cancel
out, since they traverse the same path except in opposite directions. Hence
Z
XZ
(D.10)
g dy = g dy,
j
j

and (D.3) follows.

A garden variety piecewise smoothly bounded domain might not be of type I or type
II, but typically can be divided into domains j of type I, as depicted in Fig. D.4. For
each such , we have
Z
ZZ
f
(D.11)
f dx =
dx dy,
y
j

and summing over j yields (D.2). Meanwhile, one can typically divide into domains of
type II, as depicted in Fig. D.5, get (D.8) on each such domain, and sum over j to get
(D.3).

260

It is possible to concoct piecewise smoothly bounded domains that would require an

infinite number of divisions to yield subdomins of type I (or of type II). In such a case
a limiting argument can be used to establish (D.1). We will not dscuss the details here.
Arguments applying to general domains can be found in [T3], in Appendix G for domains
with C 2 boundary, and in Appendix I for domains substantially rougher than treated here.
Of use for the proof of Cauchys integral theorem in 5 is the special case f = iu, g = u
of (D.1), which yields

(D.12)

ZZ
Z
u
u
+i
dx dy = i u (dx + i dy)
x
y

Z
= i u dz.

When u C 1 () is holomorphic in , the integrand on the left side of (D.12) vanishes.

Another special case arises by taking f = iu, g = u. Then we get

(D.13)

ZZ
Z
u
u
i
dx dy = i u (dx i dy)
x
y

Z
= i u dz.

In case u is holomorphic and u0 (z) = v(z), the integrand on the left side of (D.13) is 2v(z).
Such an identity (together with a further limiting argument) is useful in Exercise 7 of 30.

261

E. Poincar
e metrics
Recall from 22 that the upper half plane
(E.1)

U = {z C : Im z > 0}

az + b
a b
(E.2)
LA (z) =
, A=
,
c d
cz + d
of the form
(E.3)

while the disk

(E.4)

D = {z C : |z| < 1}

is invariant under the group of transformations LB with

a b
(E.5)
B SU (1, 1), i.e., B =
, |a|2 |b|2 = 1.
b a
These groups are related by the holomorphic diffeomorphism

zi
1 i
(E.6)
LA0 = : U D, A0 =
, i.e., (z) =
.
1 i
z+i
Here we produce metric tensors on U and D that are invariant under these respective group
actions. These are called Poincare metrics.
We start with U, and the metric tensor
(E.7)

gjk (z) =

1
jk ,
y2

(E.8)

dsU =

1
|dz|.
y

(E.9)

(z) = z + , R,

r (z) = rz, r (0, ).

262

Note that

(E.10)

= LT , r = LDr ,

T =

1
0 1

, Dr =

r1/2
0

0
r1/2

The group generated by these operations is not all of Sl(2, R). We proceed as follows.
We pull the metric tensor (E.7) back to the disk D, via = 1 , given by
(E.11)

(z) =

1z+1
,
i z1

: D U.

We have
(E.12)

dz = 0 (z) dz =

2i
dz,
(z 1)2

and hence
1
| dz|
Im (z)
2i
| 0 (z)| |dz|
=
(z) (z)
4
|dz|
= z+1 z+1
(z 1)(z 1)
z1 + z1
2
=
|dz|.
1 |z|2

dsU =

(E.13)

(E.14)

dsD =

2
|dz|,
1 |z|2

or
(E.15)

hjk (z) =

4
jk .
(1 |z|2 )2

This metric tensor is invariant under

(E.16)

LTe , LD
er ,

1
e
Te = A0 T A1
0 , Dr = A0 Dr A0 .

(E.17)

(z) = ei z,

R.

263

Here

(E.18)

= LR ,

R =

ei/2
0

ei/2

e r and R can be seen to generate all of SU (1, 1), which implies

The transformations Te , D
the metric tensor (E.11) on D is invariant under all the linear fractional transformations
(E.5), and hence the metric tensor (E.7) on U is invariant under all the linear fractional
transformations (E.2)(E.3). Alternatively, one can check directly that
a,b dsD = dsD ,

(E.19)
when
(E.20)

a,b (z) =

az + b
,
bz + a

|a|2 |b|2 = 1.

In fact,
(E.21)

a,b dz = 0a,b (z) dz =

dz
,
(bz + a)2

and hence
2
|0 (z)| |dz|
1 |a,b (z)|2 a,b
2 |dz|
=
|bz + a|2 |az + b|2
2 |dz|
=
1 |z|2
= dsD .

a,b dsD =
(E.22)

Let us record the result formally.

Proposition E.1. The metric tensor (E.15) on D is invariant under all linear fractional
transformations of the form (E.20). Hence the metric tensor (E.7) on U is invariant under
all linear fractional transformations of the form (E.2)(E.3).
These metric tensors are called Poincare metrics, and U and D, equipped with these
metric tensors, are called the Poincare upper half plane and the Poincare disk.
Let C be a simply connected domain, 6= C. The Riemann mapping theorem
produces a holomorphic diffeomorphism
(E.23)

: D.

The pull back of the Poincare metric on D via is called the Poincare metric on . Note
e : D is another holomorphic diffeomorphism, then
e 1 : D D is a
that if

264

holomorphic diffeomorphism, hence, as seen in 22, a linear fractional transformation of

e pull back the
the form (E.20), hence it preserves the Poincare metric on D, so and
Poincare metric to the same metric tensor on .
More generally, a domain C inherits a Poincare metric whenever there is a holomorphic covering map
(E.24)

: D .

In fact, for each q , if you choose p 1 (q), is a holomorphic diffeomorphism from

a neighborhood Op of p onto a neighborhood Oq of q, and the Poincare metric on D pulls
back via 1 : Oq Op . This is independent of the choice of p 1 (q), since two
such inverses 1 and 1 : Oq Op0 are related by a covering map on D, which must
be of the form (E.20). For the same reason, any other covering map D produces
the same metric on , so one has a well defined Poincare metric on , whenever there is
a holomorphic covering map (E.24). Such a metric tensor is always a multiple of jk in
standard (x, y)-coordinates,
(E.25)

or
(E.26)

ds = A (z) |dz|,

inverse q to ,
(E.27)

with AD given by (E.14), i.e.,

(E.28)

2
.
1 |z|2

The following is a definitive result on the class of domains to which such a construction
applies.
Theorem E.2. If C is a connected open set and C \ contains at least two points,
then there is a holomorphic covering map (E.24).
This is part of the celebrated Uniformization Theorem, of which one can read a careful
account in [For]. We will not give a proof of Theorem E.2 here. A proof using basic results
about partial differential equations is given in [MaT]. We recall that 25 establishes this
result for = C \ {0, 1}. To see how Theorem E.2 works for D = D \ {0}, note that
(E.29)

: U D ,

(z) = eiz ,

is a holomorphic covering map, and composing with the inverse of in (E.6) yields a
holomorphic covering map D D .
The following interesting result is a geometric version of the Schwarz lemma.

265

Proposition E.3. Assume O and are domains in C with Poincare metrics, inherited
from holomorphic coverings by D, and F : O is holomorphic. Then F is distancedecreasing.
What is meant by distance decreasing is the following. Let : [a, b] O be a smooth
path. Its length, with respect to the Poincare metric on O, is
Z
(E.30)

`O () =

AO ((s))| 0 (s)| ds.

The assertion that F is distance decreasing is that, for all such paths ,
(E.31)

` (F ) `O ().

Rb
Note that ` (F ) = a A (F (s))|(F )0 (s)| ds and (F )(s) = F 0 ((s)) 0 (s), so
the validity of (E.31) for all paths is equivalent to
(E.32)

A (F (z))|F 0 (z)| AO (z),

z O.

To prove Proposition E.3, note that under the hypotheses given there, F : O lifts
to a holomorphic map G : D D, and it suffices to show that any such G is distance
decreasing, for the Poincare metric on D, i.e.,
(E.33)

G : D D holomorphic = AD (G(z0 ))|G0 (z0 )| AD (z0 ),

z0 D.

Now, given z0 D, we can compose G on the right with a linear fractional transformation
of the form (E.20), taking 0 to z0 , and on the left by such a linear fractional transformation,
taking G(z0 ) to 0, obtaining
(E.34)

H : D D holomorphic, H(0) = 0,

(E.35)

|H 0 (0)| 1,

(E.36)

|H(z)| |z|.

This in turn is the conclusion of the Schwarz lemma, Proposition 6.2.

Using these results, we will give another proof of Picards big theorem, Proposition 25.2.

266

Proposition E.4. If f : D C\{0, 1} is holomorphic, then the singularity at 0 is either

a pole or a removable singularity.
To start, we assume 0 is not a pole or removable singularity, and apply the CasoratiWeierstrass theorem, Proposition 11.3, to deduce the existence of aj , bj D such that
(E.37)

pj = f (aj ) 0,

qj = f (bj ) 1,

as j . Necessarily aj , bj 0. Let j be the circle centered at 0 of radius |aj | and j

the circle centered at 0 of radius |bj |. An examination of (E.29) reveals that
(E.38)

`D (j ) 0,

`D (j ) 0.

(E.39)

j = f j ,

j = f j

(E.40)

`C (
j ) 0,

`C (
j ) 0.

We now bring in the following:

Lemma E.5. Fix z0 C = C \ {0, 1}. Let j be paths from z0 to pj 0. Then
`C (j ) . A parallel result holds for paths from z0 to qj 1.
Proof. Let : D C be a holomorphic covering, z0 1 (z0 ), and j a lift of j to a
path in D starting at z0 . Then
(E.41)

`C (j ) = `D (
j ).

Now j contains points that tend to D (in the Euclidean metric) as j , so the fact
that `D (
j ) follows easily from the formula (E.14).
Returning to the setting of (E.37)(E.40), we deduce that
Given > 0, N < such than whenever j N
(E.42)

f j = j {z C : |z| < }, and

f j =
j {z C : |z 1| < }.

If, e.g., < 1/4, this cannot hold for a function f that is holomorphic on a neighborhood
of the annular region bounded by j and j . (Cf. 6, Exercise 2.) Thus our assumption
on f contradicts the Casorati-Weierstrass theorem, and Proposition E.4 is proven.

267

F. The fundamental theorem of algebra (elementary proof )

Here we provide a proof of the fundamental theorem of algebra that is quite a bit
different from that given in 6. The proof here is more elementary, in the sense that it
does not make use of results from integral calculus. On the other hand, it is longer than
the proof from 6. Here it is.
Theorem F.1. If p(z) is a nonconstant polynomial (with complex coefficients), then p(z)
must have a complex root.
Proof. We have, for some n 1, an 6= 0,
(F.1)

p(z) = an z n + + a1 z + a0

= an z n 1 + O(z 1 ) , |z| ,

which implies
(F.2)

lim |p(z)| = .

|z|

(F.3)

|z|R

we deduce that
(F.4)

|z|R

zC

(F.5)

zC

The theorem hence follows from:

Lemma F.2. If p(z) is a nonconstant polynomial and (F.5) holds, then p(z0 ) = 0.
Proof. Suppose to the contrary that
(F.6)

p(z0 ) = a 6= 0.

We can write
(F.7)

p(z0 + ) = a + q(),

268

where q() is a (nonconstant) polynomial in , satisfying q(0) = 0. Hence, for some k 1

and b 6= 0, we have q() = b k + + bn n , i.e.,
(F.8)

q() = b k + O( k+1 ),

0,

so, uniformly on S 1 = { : || = 1}
(F.9)

p(z0 + ) = a + b k k + O(k+1 ),

& 0.

(F.10)

b k
a
= ,
|b|
|a|

which is possible since a 6= 0 and b 6= 0. Then

(F.11)

p(z0 + ) = a 1 k + O(k+1 ),
a

which contradicts (F.5) for > 0 small enough. Thus (F.6) is impossible. This proves
Lemma F.2, hence Theorem F.1.

269

G. The Weierstrass approximation theorem

In this appendix we establish the following result of Weierstrass, on the approximation
by polynomials of an arbitrary continuous function on a closed bounded interval [a, b] R.
Theorem G.1. If f : [a, b] C is continuous, then there exist polynomials pk (x) such
that pk (x) f (x) uniformly on [a, b].
For the proof, first extend f to be continuous on [a1, b+1] and vanish at the endpoints.
We leave it to the reader to do this. Then extend f to be 0 on (, a1] and on [b+1, ).
Then we have a continuous function f : R C with compact support. We write f C0 (R).
As seen in 14, if we set
2
1
(G.1)
H (x) =
ex /4 ,
4
for > 0 and form
Z
(G.2)
f (x) =
H (x y)f (y) dy,

then f (x) f (x) as 0, for each x R. In fact it folows easily that f f uniformly
on R, whenever f C0 (R). In particular, given > 0, there exists > 0 such that
(G.3)

|f (x) f (x)| < ,

x R.

Now note that, for each > 0, f (x) extends from x R to the entire holomorphic
function
Z
2
1
e(zy) /4 f (y) dy.
(G.4)
F (z) =
4
That this integral is absolutely convergent for each z C is elementary, and that it is
holomorphic in z can be deduced from Moreras theorem. It follows that F (z) has a
power series expansion,

X
(G.5)
F (z) =
an ()z n ,
n=0

(G.6)

where
(G.7)

PN, (z) =

N
X

an ()z n .

n=0

Consequently, by (G.3) and (G.6),

(G.8)
This proves Theorem G.1.

x [a, b].

270

H. Inner product spaces

On occasion, particularly in 1314, we have looked at norms and inner products on
spaces of functions, such as C(S 1 ) and S(R), which are vector spaces. Generally, a complex
vector space V is a set on which there are operations of vector addition:
(H.1)

f, g V = f + g V,

(H.2)

a C, f V = af V,

satisfying the following properties. For vector addition, we have

(H.3)

f + g = g + f, (f + g) + h = f + (g + h), f + 0 = f, f + (f ) = 0.

(H.4)

a(bf ) = (ab)f,

1 f = f.

Furthermore, we have two distributive laws:

(H.5)

a(f + g) = af + ag,

(a + b)f = af + bf.

These properties are readily verified for the function spaces arising in 1314.
An inner product on a complex vector space V assigns to elements f, g V the quantity
(f, g) C, in a fashion that obeys the following three rules:
(a1 f1 + a2 f2 , g) = a1 (f1 , g) + a2 (f2 , g),
(H.6)

(f, g) = (g, f ),
(f, f ) > 0

unless f = 0.

A vector space equipped with an inner product is called an inner product space. For
example,
Z
1
(H.7)
(f, g) =
f ()g() d
2
S1

Z
(H.8)
(f, g) =
f (x)g(x) dx

271

defines an inner product on S(R) (defined in 14). As another example, in 13 we defined

`2 to consist of sequences (ak )kZ such that

(H.9)

|ak |2 < .

k=

(H.10)

(ak ), (bk ) =
ak bk .
k=

Given an inner product on V , one says the object kf k defined by

(H.11)

kf k =

(f, f )

is the norm on V associated with the inner product. Generally, a norm on V is a function
f 7 kf k satisfying
(H.12)
(H.13)
(H.14)

kaf k = |a| kf k,
kf k > 0

a C, f V,

unless f = 0,

kf + gk kf k + kgk.

The property (H.14) is called the triangle inequality. A vector space equipped with a norm
is called a normed vector space. We can define a distance function on such a space by
(H.15)

d(f, g) = kf gk.

Properties (H.12)(H.14) imply that d : V V [0, ) satisfies the properties in (A.1),

making V a metric space.
If kf k is given by (H.11), from an inner product satisfying (H.6), it is clear that (H.12)
(H.13) hold, but (H.14) requires a demonstration. Note that
kf + gk2 = (f + g, f + g)
(H.16)

= kf k2 + (f, g) + (g, f ) + kgk2

= kf k2 + 2 Re(f, g) + kgk2 ,

while
(H.17)

(kf k + kgk)2 = kf k2 + 2kf k kgk + kgk2 .

Thus to establish (H.17) it suffices to prove the following, known as Cauchys inequality.

272

Proposition H.1. For any inner product on a vector space V , with kf k defined by (H.11),
(H.18)

f, g V.

(H.19)

0 kf gk2 = kf k2 2 Re(f, g) + kgk2 ,

which implies
(H.20)

2 Re(f, g) kf k2 + kgk2 ,

f, g V.

Replacing f by af for arbitrary a C of absolute velue 1 yields 2 Re a(f, g) kf k2 + kgk2 ,

for all such a, hence
(H.21)

f, g V.

(H.22)

2|(f, g)| t2 kf k2 + t2 kgk2 ,

f, g V, t (0, ).

If we take t2 = kgk/kf k, we obtain the desired inequality (H.18). This assumes f and g
are both nonzero, but (H.18) is trivial if f or g is 0.
An inner product space V is called a Hilbert space if it is a complete metric space, i.e.,
if every Cauchy sequence (f ) in V has a limit in V . The space `2 has this completeness
property, but C(S 1 ), with inner product (H.7), does not. Appendix A describes a process
of constructing the completion of a metric space. When appied to an incomplete inner
product space, it produces a Hilbert space. When this process is applied to C(S 1 ), the
completion is the space L2 (S 1 ), briefly discussed in 13. This result is essentially the
content of Propositions A and B, stated in 13.
There is a great deal more to be said about Hilbert space theory, but further material
is not needed here. One can consult a book on functional analysis, or the appendix on
functional analysis in Vol. 1 of [T2].

273

I. 2 is Irrational
The following proof that 2 is irrational follows a classic argument of D. Niven. The
idea is to consider
Z
1
(I.1)
In =
n (x) sin x dx, n (x) = xn ( x)n .
n!
0
Clearly In > 0 for each n N, and In 0 very fast, faster than geometrically. The next
key fact is that In is a polynomial of degree n in 2 with integer coefficients:
(I.2)

In =

n
X

cnk 2k ,

cnk Z.

k=0

(3)

n
X

cnk a2k b2n2k = b2n In .

k=0

But the left side of (I.3) is an integer for each n, while by the estimate on (I.1) mentioned
above the right side belongs to the interval (0, 1) for large n, yielding a contradiction. It
remains to establish (I.2).
A method of computing the integral in (I.1), which works for any polynomial n (x)) is
the following. One looks for an antiderivative of the form
(I.4)

(I.5)

hence
(I.6)

Fn00 (x) + Fn (x) = n (x).

One can exploit the nilpotence of x2 on the space of polynomials of degree 2n and set
Fn (x) = (I + x2 )1 n (x)
(I.7)

n
X
k=0

(1)k (2k)
n (x).

274

Then

d 0
Fn (x) sin x Fn (x) cos x = n (x) sin x.
dx

(I.8)

Z

(I.9)

n (x) sin x dx = Fn (0) + Fn () = 2Fn (0),

0

the last identity holding for n (x) as in (I.1) because then n ( x) = n (x) and hence
Fn ( x) = Fn (x). For the first identity in (I.9), we use the defining property that
sin = 0 while cos = 1.
(2k)
In light of (I.7), to prove (I.2) it suffices to establish an analogous property for n (0).
Comparing the binomial formula and Taylors formula for n (x):

n
1 X
` n
n` xn+` ,
(1)
n (x) =
n!
`

and

`=0

(I.10)
n (x) =

2n
X
k=0

1 (k)
n (0)xk ,
k!

we see that
(I.11)

k =n+`

(k)
n (0)

` (n

+ `)! n n`

,
n!
`

n (n

+ `)! n 2(k`)

.
n!
`

= (1)

so
(I.12)

2k = n + `
(2k)

Of course n
In fact,

(I.13)

(2k)
n (0)

= (1)

(0) = 0 for 2k < n. Clearly the multiple of 2(k`) in (I.12) is an integer.

(n + `)! n
(n + `)!
n!
=
n!
`
n!
`!(n `)!
(n + `)!
n!
=
n!`! (n `)!

n+`
=
n(n 1) (n ` + 1).
n

Thus (I.2) is established, and the proof that 2 is irrational is complete.

275

J. Eulers constant
Here we say more about Eulers constant, introduced in (18.17), in the course of producing the Euler product expansion for 1/(z). The definition
(J.1)

n
X

1
= lim
log(n + 1)
n
k
k=1

of Eulers constant involves a very slowly convergent sequence. In order to produce a

numerical approximation
of , it is convenient to use other formulas, involving the Gamma
R
function (z) = 0 et tz1 dt. Note that
Z

(J.2)

(z) =

Meanwhile the Euler product formula 1/(z) = zez

n=1 (1

+ z/n)ez/n implies

0 (1) = .

(J.3)
Thus we have the integral formula

Z
(J.4)

To evaluate this integral numerically it is convenient to split it into two pieces:

Z
(J.5)

(log t)e

dt

(log t)et dt

= a b .
We can apply integration by parts to both the integrals in (5), using et = (d/dt)(et 1)
on the first and et = (d/dt)et on the second, to obtain
Z
(J.6)

a =
0

1 et
dt,
t

b =
1

et
dt.
t

Using the power series for et and integrating term by term produces a rapidly convergent series for a :
(J.7)

a =

X
(1)k1
k=1

k k!

276

Before producing infinite series representations for b , we note that the change of variable
t = sm gives
Z sm
e
(J.8)
b = m
ds,
s
1
which is very well approximated by the integral over s [1, 10) if m = 2, for example.
To produce infinite series for b , we can break up [1, ) into intervals [k, k + 1) and
take t = s + k, to write
(J.9)

b =

X
ek
k=1

Z
k ,

k =
0

et
dt.
1 + t/k

Note that 0 < k < 1 1/e for all k. For k 2 we can write

X
1 j
k =
j ,

k
j=0

(J.10)

j =

tj et dt.

(J.11)

Ej (t) =

j
X
t`
`=0

`!

Then
(J.12)

Ej (t) = Ej1 (t),

hence
(J.13)

d
tj
Ej (t)et = Ej1 (t) Ej (t) et = et ,
dt
j!

so

Z
tj et dt = j!Ej (t)et + C.

(J.14)
In particular,
Z
j =

j t

t e
0

(J.15)

1 X 1
dt = j! 1
e
`!
j

`=0

j!
1
e
`!
`=j+1

1 1
1
=
+
+ .
e j + 1 (j + 1)(j + 2)
=

277

Z
1

1 =
1

(J.16)

et
dt
t

X
(1)j
j=0

j!

= log 2 +

tj1 dt

X
(1)j
j=1

j j!

(2j 1).

To summarize, we have = a b , with a given by the convenient series (J.7) and

(J.17)

X
X
ek 1 j
(1)j j
b =

j + log 2 +
(2 1),
k
k
j j!
j=0
j=1
k=2

with j given by (J.15). We can reverse the order of summation of the double series and
write
(J.18)

b =

(1) j j + log 2 +

j=0

X
(1)j
j=1

j j!

(2j 1).

with

X
ek
j =
.
k j+1

(J.19)

k=2

Note that
(J.20)

(j+1)

0 < j < 2

ek < 2(j+3) ,

k=2

while (J.15) readily yields 0 < j < 1/ej. So one can expect 15 digits of accuracy by
summing the first series in (J.18) over 0 j 50 and the second series over 0 j 32,
assuming the ingredients j and j are evaluated sufficiently accurately. It suffices to sum
(J.19) over 2 k 40 2j/3 to evaluate j to sufficient accuracy.
Note that the quantities j do not have to be evaluated independently. Say you are
summing the first series in (J.18) over 0 j 50. First evaluate 50 using 20 terms in
(J.15), and then evaluate inductively 49 , . . . , 0 using the identity
(J.21)

j1 =

j
1
+
,
je
j

equivalent to j = jj1 1/e, which follows by integration by parts of

R1
0

tj et dt.

278

If we sum the series (J.7) for a over 1 k 20 and either sum the series (J.18) as
described above or have Mathematica numerically integrate (J.8), with m = 2, to high
precision, we obtain
(J.22)

0.577215664901533,

which is accurate to 15 digits.

We give another series for . This one is more slowly convergent than the series in (J.7)
and (J.18), but it makes clear why exceeds 1/2 by a small amount, and it has other
(J.23)

X
n=1

n ,

1
n =
n

n+1
n

1
1
dx
= log 1 +
.
x
n
n

Thus n is the area of the region

(J.24)

n
1
1o
n = (x, y) : n x n + 1,
y
.
x
n

This region contains the triangle Tn with vertices (n, 1/n), (n+1, 1/n), and (n+1, 1/(n+1)).
The region n \ Tn is a little sliver. Note that
(J.25)

Area Tn = n =

1 1
1

,
2 n n+1

and hence

(J.26)

n =

n=1

1
.
2

Thus
(J.27)

1
= (1 1 ) + (2 2 ) + (3 3 ) + .
2

Now
(J.28)

1 1 =

3
log 2,
4

(J.29)

1
1
1
3 + 4
2
2n
3n
4n
1
1
1
n = 2 3 + 4 ,
2n
2n
2n

n =

279

(J.30)

n =

1
1
1
= 2
2n(n + 1)
2n 1 +

1
n

(J.31)

1 1 X 1
1 1 X 1
1
= (1 1 ) +

+ ,
2
2 3
n3
2 4
n4
n2

n2

or, with
(k) =

(J.32)

X 1
,
nk

n1

we have
(J.33)

1 1
1 1
1 3
=
log 2 +

[(3) 1]
[(4) 1] + ,
2
4
2 3
2 4

an alternating series from the third term on. We note that

3
log 2 0.0568528,
4

(J.34)

1
[(3) 1] 0.0336762,
6
1
[(4) 1] 0.0205808,
4
3
[(5) 1] 0.0110783.
10

The estimate
(J.35)

Z
X 1
k
<2 +
xk dx
nk
2

n2

implies
(J.36)

1
0<

[(k) 1] < 2k ,
2 k

so the series (J.33) is geometrically convergent. If k is even, (k) is a known rational

multiple of k . However, for odd k, the values of (k) are more mysterious. Note that to
get (3) to 16 digits by summing (J.32) one needs to sum over 1 n 108 . On a 1.3 GHz
personal computer, a C program does this in 4 seconds. Of course, this is vastly slower
than summing (J.7) and (J.18) over the ranges discussed above.

280

K. Rapid evaluation of the Weierstrass -function

Given a lattice C, the associated Weierstrass -function is defined by
(K.1)

X
1
1
1
(z; ) = 2 +
2 .
z
(z )2

06=

This converges rather slowly, so another method must be used to evaluate (z; ) rapidly.
The classical method, which we describe below, involves a representation of in terms of
theta functions. It is most conveniently described in case
(K.2)

generated by 1 and ,

Im > 0.

To pass from this to the general case, we can use the identity
(K.3)

(z; a) =

1 z
; .
a2
a

The material below is basically a summary of material from 32, assembled here to clarify
the important application to the task of the rapid evaluation of (K.1).
To evaluate (z; ), which we henceforth denote (z), we use the following identity:
(K.4)

0 (0) (z) 2
2
1
(z) = e1 +
.
2 (0) 1 (z)

See (32.20). Here e1 = (1 /2) = (1/2), and the theta functions j (z) (which also
depend on ) are defined as follows (cf. (32.6)(32.10)):
1 (z) = i
2 (z) =
(K.5)
3 (z) =
4 (z) =

(1)n p2n1 q (n1/2) ,

n=

p2n1 q (n1/2) ,

n=

X
n=

p2n q n ,
2

(1)n p2n q n .

n=

Here
(K.6)

p = eiz ,

q = ei ,

281

with as in (K.2).
The functions 1 and 2 appear in (K.4). Also 3 and 4 arise to yield a rapid evaluation
of e1 (cf. (32.33)):
(K.7)

e1 =

2
3 (0)4 + 4 (0)4 .
3

01 (0) =

(K.8)

n=

1 (z) = i
(K.9)
2 (z) =

n=1

q (n1/2) (p2n1 + p12n ).

n=1

also formulas for 01 (0) and j (0), which appear in (K.4) and (K.7), can be rewritten:
01 (0)

= 2

2 (0) = 2
(K.10)

(1)n (2n 1)q (n1/2) ,

n=1

X
(n1/2)2

n=1

3 (0) = 1 + 2
4 (0) = 1 + 2

qn ,

n=1

(1)n q n .

n=1

Rectangular lattices
We specialize to the case where is a rectangular lattice, of sides 1 and L, more
precisely:
(K.11)

L > 0.

Now the formulas established above hold, with = iL, hence

(K.12)

q = eL .

282

Since q is real, we see that the quantities 01 (0) and j (0) in (K.10) are real. It is also
convenient to calculate the real and imaginary parts of j (z) in this case. Say
(K.13)

z = u + iv,

p2n1 = e(2n1)v cos(2n 1)u + i sin(2n 1)u .

(K.14)
We then have

Re(i1 (z)) =

2
(1)n q (n1/2) e(2n1)v e(2n1)v cos(2n 1)u,

n=1

(K.15)
Im(i1 (z)) =

2
(1)n q (n1/2) e(2n1)v + e(2n1)v sin(2n 1)u,

n=1

and
Re 2 (z) =
(K.16)

2
q (n1/2) e(2n1)v + e(2n1)v cos(2n 1)u,

n=1

Im 2 (z) =

2
q (n1/2) e(2n1)v e(2n1)v sin(2n 1)u.

n=1

We can calculate these quantities accurately by summing over a small range. Let us
insist that
(K.17)

1
1
u< ,
2
2

L
L
v< ,
2
2

and assume
(K.18)

L 1.

Then
(K.19)

(n1/2)2 (2n1)v
q
e(n2 3n+5/4)L ,
e

and since
e <

(K.20)

1
,
20

(K.21)

< 0.5 1014

for n = 5,

25

for n = 6,

< 2 10

283

with rapid decrease for n > 6. Thus, summing over 1 n 5 will give adequate approximations.
For z = u + iv very near 0, where 1 vanishes and has a pole, the identity
(0) (z) 2
1
2
1
=
,
(z) e1
01 (0) 2 (z)

(K.22)

in concert with (K.10) and (K.15)(K.16), gives an accurate approximation to ((z)e1 )1 ,

which in this case is also very small. Note, however, that some care should be taken in
evaluating Re(i1 (z)), via the first part of (K.15), when |z| is very small. More precisely,
care is needed in evaluating
ekv ekv ,

(K.23)

k = 2n 1 {1, 3, 5, 7, 9},

when v is very small, since then (K.23) is the difference between two quantities close to 1,
so evaluating ekv and ekv separately and subtracting can lead to an undesirable loss of
accuracy. In case k = 1, one can effect this cancellation at the power series level and write
(K.24)

=2

j1,odd

(v)j
.
j!

If |v| 102 , summing over j 7 yields substantial accuracy. (If |v| > 102 , separate
evaluation of ekv and ekv should not be a problem.) For other values of k in (K.23),
one can derive from
(xk 1) = (x 1)(xk1 + + 1)

(K.25)
the identity
(K.26)

kv

kv

= (e

k1
X

e(2`(k1))v ,

`=0

which in concert with (K.24) yields an accurate evaluation of each term in (K.23).
Remark. If (K.11) holds with 0 < L < 1, one can use (K.3), with a = iL, to transfer to
the case of a lattice generated by 1 and i/L.

284

References

[Ahl]
[BN]
[BB]
[CG]
[Cl]
[Con]
[Ed]
[FK]
[For]
[Gr]
[Hil]
[Kra]
[Law]
[MaT]
[MM]
[Mil]
[Mun]
[Sch]
[Sp]
[T0]
[T]
[T2]
[T3]
[Ts]

L. Ahlfors, Complex Analysis, McGraw-Hill, New York, 1966.

J. Bak and D.J. Newman, Complex Analysis, Springer-Verlag, New York, 1982.
J. Borwein and P. Borwein, Pi and the AGM, Wiley, New York, 1987.
L. Carleson and T. Gamelin, Complex Dynamics, Springer-Verlag, New York, 1993.
H. Clemens, A Scrapbook of Complex Curves, Plenum, New York, 1980.
J. Conway, Functions of One Complex Variable, Springer-Verlag, New York, 1978.
H. Edwards, Riemanns Zeta Function, Dover, New York, 2001.
H. Farkas and I. Kra, Riemann Surfaces, Springer-Verlag, New York, 1980.
O. Forster, Lectures on Riemann Surfaces, Springer-Verlag, New York, 1981.
M. Greenberg, Lectures on Algebraic Topology, Benjamin, Reading, Mass., 1967.
E. Hille, Analytic Function Theory, Vols. 12, Chelsea, New York, 1962.
S. Krantz, Complex Analysis: The Geometric Viewpoint, Carus Math. Monogr.
#23, Mathematical Association of America, 1990.
D. Lawden, Elliptic Functions and Applications, Springer-Verlag, New York, 1989.
R. Mazzeo and M. Taylor, Curvature and Uniformization, Israel J. Math. 130
(2002), 179196.
H. McKean and V. Moll, Elliptic Curves, Cambridge Univ. Press, 1999.
J. Milnor, Dynamics in One Complex Variable, Vieweg, Braunsweig, 1999.
J. Munkres, Topology, a First Course, Prentice-Hall, Englewood Cliffs, New Jersey,
1975.
J. Schiff, Normal Families, Springer-Verlag, New York, 1993.
M. Spivak, A Comprehensive Introduction to Differential Geometry, Vols. 15, Publish or Perish Press, Berkeley, Calif., 1979.
M. Taylor, Introduction to Analysis in One Variable, Lecture notes, available at
http://www.unc.edu/math/Faculty/met/math521.html
M. Taylor, Introduction to Analysis in Several Variables, Lecture notes, available
at http://www.unc.edu/math/Faculty/met/math521.html
M. Taylor, Partial Differential Equations, Vols. 13, Springer-Verlag, New York,
1996 (2nd ed. 2011).
M. Taylor, Measure Theory and Integration, American Mathematical Society, Providence RI, 2006.
M. Tsuji, Potential Theory and Modern Function Theory, Chelsea, New York, 1975.

285

Index
Abel inversion problem, 217
analytic continuation, 85
argument principle, 135
arc length, 43
Arzela-Ascoli theorem, 160, 240
Arithmetic-geometric mean, 223
Bernoulli numbers, 97, 201
binomial coefficient, 37, 40, 134
branched covering, 227
Casorati-Weierstrass theorem, 90
Cauchy integral formula, 56
Cauchy integral theorem, 56
Cauchy-Riemann equation, 24
Cauchys inequalities, 62, 108
Cauchy sequence 10, 233
chain rule, 20, 26, 245
compact space, 233
complex differentiable, 20
conformal map, 162
connected, 242
contraction mapping theorem, 242
convolution, 111, 120, 128
cos 43
covering map, 158, 175
derivative, 13, 20, 245
Dirichlet problem, 109, 193
domain
elliptic function, 199, 205
elliptic integral, 206, 217
of first, second, third kind, 220
essential singularity, 90
Eulers constant, 143, 275
Eulers formula, 43
Euler product formula, 143
exponential function, 15, 21, 39

286

Fatou set, 187

Fourier inversion formula, 114
Fourier series, 98
Fourier transform, 113
fundamental theorem of algebra, 66, 138, 267
fundamental theorem of calculus 13
Gamma function, 141
Gaussian integral, 85, 144
geometric series, 10
Goursats theorem, 83
Greens theorem, 56, 258
harmonic conjugate, 70
harmonic function, 70
Harnack estimate, 193
holomorphic, 20
Hurwitz theorem, 138
infinite product, 149
inverse function, 48, 247
Jacobi identity, 120, 154
Julia set, 187
Laplace transform, 125
Laurent series, 93
Legendre duplication formula, 149
lift, 158
line integral, 26
linear fractional transformation, 162
Liouvilles theorem, 65, 193
log, 41, 49
maximum principle, 64, 73
mean value property, 57, 64, 72
meromorphic function, 90
metric space, 233
Montels theorem, 160, 186
Moreras theorem, 80
normal family, 160

287

open mapping theorem, 138

path integral, 26
pi, 42
Picards theorems, 191
Poincare disk, 263
Poincare metric, 261
Poisson integral, 109
Poisson summation formula, 120
polynomial, 21, 66
pole, 90
power series, 10, 31
radius of convergence, 12, 31
removable singularity, 89
residue, 129
Riemann functional equation, 155
Riemann mapping theorem, 168
Riemann sphere, 179
Riemann surface, 181, 226
Riemann zeta function, 153
Rouches theorem, 137
Schwarz lemma, 65
Schwarz reflection principle, 80, 109
sec, 45
simply connected, 63, 74, 158, 168
sin, 43
spherical derivative, 183
star shaped 60
surface, 250
tan, 46
theta function, 211
trigonometric functions, 39
torus, 181, 201
Weierstrass approximation, 269
Weierstrass -function, 199, 205
winding number, 136